FooCars_Step_By_Step.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Copy of Copy of  FooCars Step By Step",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/lcipolina/foocars/blob/lucia/FooCars_Step_By_Step.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i5ejv-Jt8D__",
        "colab_type": "text"
      },
      "source": [
        "# What is Foocars and why should I care?\n",
        "\n",
        "Foocars is an inexpensive open source autonomous car driving system. It uses machine learning to create a robust method of driving cars around cones and obstaces.  \"Cars\" is broad term in this case. It's been used on toy cars, RC Cars, and single seater Power Wheels vehicles. This is an extremely low cost and robust option to get started in understanding autonomous vehicles.\n",
        "\n",
        "For this project we''ll use the RC car version and work out how the car trains and later makes decisions based on the collected training data.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V0tGN0AY84dj",
        "colab_type": "text"
      },
      "source": [
        "# The Foocars project \n",
        "\n",
        "It's located here: https://foocars.io\n",
        "\n",
        "The main working repository for the project is here: https://github.com/fubarlabs/foocars"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ugfhzhg35_PD",
        "colab_type": "text"
      },
      "source": [
        "# The Code\n",
        "This code implements Machine Learning techniques to generate an algorithm that allows a car to take driving decisions autonomously. \n",
        "The  algorithm is capable of identifying obstacles on a racing circuit allowing the car to avoid them.  \n",
        "\n",
        "To calibrate (i.e. train) the algorithm we use pre-recorded images of obstacles usually found on a racing circuit. We process the images using Convolutional Neuronal Networks and obtain a set of parameters that are used by the car in real time to take steering decisions while racing. \n",
        "\n",
        "The code is divided in the following sections:\n",
        "\n",
        "* Uploading the image data from an external repository\n",
        "* Training the Convolutional Neuronal Networks using Python.\n",
        "* Analizing the results\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MeFu_EZvEW7I",
        "colab_type": "text"
      },
      "source": [
        "# Uploading the Training Data\n",
        "The aim of this code is to train a Convolutional Neuronal Network to identify obtacles. We present the Network with a set of example objects and through iteration, it generates al algorithm to identify patters in the data. This allows us to classify objects into different kind of obstacles to be avoided (i.er. a cone, a shoe , a chair, etc).\n",
        "\n",
        "We store the collected data in an external repository. The most convenient way of sharing data is to create a public folder in Google Drive.\n",
        "\n",
        "This part of the code does the following:\n",
        "* Uses pipenv to create a Virtual Environment.\n",
        "* Downloads the image data and extracts it into a dedicated folder.\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GMdXz7vZQpV8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from google.colab import files"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_gEJpmY0GGMz",
        "colab_type": "text"
      },
      "source": [
        "**NOTE:** Change the below to point the global variables to where the image data is located"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O40jGjx6do7Y",
        "colab_type": "code",
        "outputId": "35d3a9e6-ecb7-47d9-8675-a8b95a7c3abe",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 69
        }
      },
      "source": [
        "%env USERNAME=lcipolina\n",
        "!echo $USERNAME\n",
        "!echo https://github.com/$USERNAME/foocars.git\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "env: USERNAME=lcipolina\n",
            "lcipolina\n",
            "https://github.com/lcipolina/foocars.git\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zRiHlWWiGXp6",
        "colab_type": "text"
      },
      "source": [
        "We will make a local copy  (i.e. a clone) of the data repository. \n",
        "Previous to this, please create your own Fork of the project's main Repo in GitHub\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JUMSIJwy9KB1",
        "colab_type": "code",
        "outputId": "95085b39-aa12-4358-865d-118ea9fed03a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# Cloning our own Fork of the project to our local repo.\n",
        "!git clone https://github.com/$USERNAME/foocars.git"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "fatal: destination path 'foocars' already exists and is not an empty directory.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5n2KtTT19Uwb",
        "colab_type": "code",
        "outputId": "9d3d570e-7f9d-4178-e188-7ea745ffa484",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "!ls"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "adc.json  data\tfoocars  sample_data\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "B53ESA5-9b6A",
        "colab_type": "code",
        "outputId": "3c4be622-90ad-4a41-aca6-8ee3757dca9c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "#we will install pipenv in foocars directory\n",
        "%cd foocars"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/foocars\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ir42Fo1ugl4L",
        "colab_type": "code",
        "outputId": "1eccf50b-8b09-42f5-8bbd-29a6f928cd39",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 69
        }
      },
      "source": [
        "!ls\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "cars\t\t  installation.md  README.md\t   tests\n",
            "Dockerfile-car\t  Pipfile\t   schematics\t   training\n",
            "Dockerfile-train  Pipfile.lock\t   setuppython.sh  utilities\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YJ1NzgqW9ihz",
        "colab_type": "code",
        "outputId": "5ab50c05-1664-4a01-fbec-af540d22a105",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 141
        }
      },
      "source": [
        "#install pipenv in \"foocars\" directory\n",
        "!pip install pipenv"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: pipenv in /usr/local/lib/python3.6/dist-packages (2018.11.26)\n",
            "Requirement already satisfied: setuptools>=36.2.1 in /usr/local/lib/python3.6/dist-packages (from pipenv) (41.0.1)\n",
            "Requirement already satisfied: virtualenv in /usr/local/lib/python3.6/dist-packages (from pipenv) (16.6.0)\n",
            "Requirement already satisfied: certifi in /usr/local/lib/python3.6/dist-packages (from pipenv) (2019.3.9)\n",
            "Requirement already satisfied: virtualenv-clone>=0.2.5 in /usr/local/lib/python3.6/dist-packages (from pipenv) (0.5.3)\n",
            "Requirement already satisfied: pip>=9.0.1 in /usr/local/lib/python3.6/dist-packages (from pipenv) (19.1.1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9aK-nROT9tBC",
        "colab_type": "code",
        "outputId": "136fff01-107b-420c-be7c-7b8651759fc9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 159
        }
      },
      "source": [
        "#Create virtual environment in the folder we are in \n",
        "!pipenv install "
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[31m\u001b[1mWarning\u001b[39m\u001b[22m: Your Pipfile requires \u001b[39m\u001b[1mpython_version\u001b[39m\u001b[22m \u001b[34m\u001b[22m3.*\u001b[39m\u001b[22m, but you are using \u001b[34m\u001b[22m3.6.7\u001b[39m\u001b[22m (\u001b[32m\u001b[22m/root/.local/share/v/f/bin/python\u001b[39m\u001b[22m).\n",
            "  \u001b[32m\u001b[22m$ pipenv --rm\u001b[39m\u001b[22m and rebuilding the virtual environment may resolve the issue.\n",
            "  \u001b[31m\u001b[22m$ pipenv check\u001b[39m\u001b[22m will surely fail.\n",
            "\u001b[39m\u001b[1mInstalling dependencies from Pipfile.lock (426ee3)…\u001b[39m\u001b[22m\n",
            "  🐍   \u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m 27/27 — \u001b[30m\u001b[22m00:00:08\u001b[39m\u001b[22m\n",
            "To activate this project's virtualenv, run \u001b[31m\u001b[22mpipenv shell\u001b[39m\u001b[22m.\n",
            "Alternatively, run a command inside the virtualenv with \u001b[31m\u001b[22mpipenv run\u001b[39m\u001b[22m.\n",
            "\u001b[0m"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4zJCnLtRw94V",
        "colab_type": "code",
        "outputId": "9ffe9729-1ec1-4ce7-a585-c484eb6f6652",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 69
        }
      },
      "source": [
        "!ls"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "cars\t\t  installation.md  README.md\t   tests\n",
            "Dockerfile-car\t  Pipfile\t   schematics\t   training\n",
            "Dockerfile-train  Pipfile.lock\t   setuppython.sh  utilities\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OcBiq0QU9pAZ",
        "colab_type": "code",
        "outputId": "be204aa4-9f5d-4a2a-b350-c34fcf45e714",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "#Create folder /steelroses\n",
        "#Download data file in folder  /content/data/cars/steelroses/collected\n",
        "!mkdir -p /content/data/cars/steelroses\n",
        "%cd /content/data/cars/steelroses"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/data/cars/steelroses\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b5mwdLnDCJWb",
        "colab_type": "code",
        "outputId": "282b1923-4f85-4882-8004-ab33b6da8cc8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# Load data from a shared folder in Google Drive\n",
        "# Tutorial: https://towardsdatascience.com/3-ways-to-load-csv-files-into-colab-7c14fcbdcb92\n",
        "\n",
        "import pdb #used to debug\n",
        "import tarfile\n",
        "\n",
        "fileID='https://drive.google.com/open?id=1jtW5aSS6dj1DNmxr87iHf1bzkBB1A3Km' # The shareable link to steelroses data\n",
        "file = 'steelroses.collected.tar.gz'\n",
        "\n",
        "# Code to read csv file into Colaboratory:\n",
        "!pip install -U -q PyDrive\n",
        "from pydrive.auth  import GoogleAuth\n",
        "from pydrive.drive import GoogleDrive\n",
        "from google.colab  import auth\n",
        "from oauth2client.client import GoogleCredentials\n",
        "# Authenticate and create the PyDrive client.\n",
        "auth.authenticate_user()\n",
        "gauth             = GoogleAuth()\n",
        "gauth.credentials = GoogleCredentials.get_application_default()\n",
        "drive             = GoogleDrive(gauth)\n",
        "link              = fileID\n",
        "fluff, id         = link.split('=')\n",
        "print (id) # Verify that you have everything after '='\n",
        "downloaded        = drive.CreateFile({'id':id}) \n",
        "\n",
        "downloaded.GetContentFile('steelroses.collected.tar.gz') \n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1jtW5aSS6dj1DNmxr87iHf1bzkBB1A3Km\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E30mQKMDy8Bx",
        "colab_type": "code",
        "outputId": "29720d1c-5564-429b-f5a2-23e52d1fdfc0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "#Extracts the data file in the folder:  /content/data/cars/steelroses/collected\n",
        "%cd /content/data/cars/steelroses\n",
        "#!tar xvzf steelroses.collected.tar.gz #verbose\n",
        "!tar xzf steelroses.collected.tar.gz #not verbose"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/data/cars/steelroses\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EjSs9u6-HJ4i",
        "colab_type": "code",
        "outputId": "ffb7585e-7784-4ef1-8700-01c26db863d8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "!ls"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "collected  steelroses.collected.tar.gz\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kSEd47PrsSHV",
        "colab_type": "text"
      },
      "source": [
        "# Environment is setup\n",
        "\n",
        "We can run the software for training. But first we need to map the data to be accessible.\n",
        "\n",
        " 1. Get g drive access to a cars curated training data\n",
        " 2. Take a look at some of the images\n",
        " 3. Train on the data set\n",
        " 4. Select steerstats \n",
        " 5. Experiement let's look at steer stats\n",
        " 6. What happens when we give it an image, what prediction is made\n",
        " 7. Download the weights and steerstats to the car\n",
        " 8. Test by running on the car"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rEGSnMWUIxRa",
        "colab_type": "text"
      },
      "source": [
        "# Description of a Convolutional Neuronal Network\n",
        "\n",
        "A Convolutional Neuronal Network is a common type of Neuronal Network used in image processing. The idea is to apply image filters in a smart way to reduce the size of images and make them more tractable. Additionally, by removing irrelevant data, the object's main features are highlighted, improving the accuracy of the object recognition algorithm. \n",
        "\n",
        "The smart filters are just mathematical operations applied in secuential order to reduce dimensions without losing accuracy.\n",
        "\n",
        "After the convolution has been applied, the resulting images are passed to the traditional Neuronal Networks to train the classification parameters.\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3VCQ40OXIARP",
        "colab_type": "text"
      },
      "source": [
        "#Train the Convolutional Neuronal Network\n",
        "\n",
        "We will use 'pipenv' to run the function inside the script 'train.py'.\n",
        "\n",
        "We need to define some variables needed by the function. Below we provide example values.:\n",
        "\n",
        "* Number of epochs:   --epochs 10 \n",
        "* --save_frequency 2 \n",
        "* path to data: \n",
        "/content/data/cars/steelroses/collected \n",
        "\n",
        "### The output of a Convolutional Neuronal Network run in Keras:\n",
        "The python code below outputs the result of the following operations:\n",
        "\n",
        "\n",
        "*  'conv2d' - Applies a 'convolution' operation (which is just a matrix by matrix multiplication) to reduce the matrix's dimension. \n",
        "\n",
        "* 'dropout'  - Is the implementation of the \"dropout technique\" to reduce parameter overfitting while preserving accuracy and classification power.\n",
        "\n",
        "* 'Flatten'  - Is used to create back a single array of data after the previous operations have been applied.\n",
        "\n",
        "These two opeations are applied iterated sequantially until a desired level or classification error is achieved."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UcF3R4ARqjGb",
        "colab_type": "code",
        "outputId": "7890658c-457e-46ec-d91d-cfcb56f103d2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "#TRAIN THE MODEL\n",
        "#Point to folder where the training function is located\n",
        "#Output will be saved to files like: weights_year_month_day_20_28_08_epoch_0.h5\n",
        "\n",
        "%cd /content/foocars/training/\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/foocars/training\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "28cJhOcRbc8R",
        "colab_type": "code",
        "outputId": "09c94314-1032-46d4-818a-35adc11c4724",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "!ls"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "defines.py\t  foocars\t    readme.md\t    train_history.py\n",
            "dropout_model.py  history_model.py  steerstats.npz  train.py\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_qHVj6LJ0PsO",
        "colab_type": "code",
        "outputId": "93ca9387-9ad5-4e03-8b34-ccaefd4c2b91",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 211
        }
      },
      "source": [
        "!pipenv install numpy"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[31m\u001b[1mWarning\u001b[39m\u001b[22m: Your Pipfile requires \u001b[39m\u001b[1mpython_version\u001b[39m\u001b[22m \u001b[34m\u001b[22m3.*\u001b[39m\u001b[22m, but you are using \u001b[34m\u001b[22m3.6.7\u001b[39m\u001b[22m (\u001b[32m\u001b[22m/root/.local/share/v/f/bin/python\u001b[39m\u001b[22m).\n",
            "  \u001b[32m\u001b[22m$ pipenv --rm\u001b[39m\u001b[22m and rebuilding the virtual environment may resolve the issue.\n",
            "  \u001b[31m\u001b[22m$ pipenv check\u001b[39m\u001b[22m will surely fail.\n",
            "\u001b[39m\u001b[1mInstalling \u001b[32m\u001b[1mnumpy\u001b[39m\u001b[22m…\u001b[39m\u001b[22m\n",
            "\u001b[K\u001b[39m\u001b[1mAdding\u001b[39m\u001b[22m \u001b[32m\u001b[1mnumpy\u001b[39m\u001b[22m \u001b[39m\u001b[1mto Pipfile's\u001b[39m\u001b[22m \u001b[31m\u001b[1m[packages]\u001b[39m\u001b[22m\u001b[39m\u001b[1m…\u001b[39m\u001b[22m\n",
            "\u001b[K\u001b[?25h✔ Installation Succeeded\u001b[0m \n",
            "\u001b[39m\u001b[1mInstalling dependencies from Pipfile.lock (426ee3)…\u001b[39m\u001b[22m\n",
            "  🐍   \u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m\u001b[32m\u001b[1m▉\u001b[39m\u001b[22m 27/27 — \u001b[30m\u001b[22m00:00:09\u001b[39m\u001b[22m\n",
            "To activate this project's virtualenv, run \u001b[31m\u001b[22mpipenv shell\u001b[39m\u001b[22m.\n",
            "Alternatively, run a command inside the virtualenv with \u001b[31m\u001b[22mpipenv run\u001b[39m\u001b[22m.\n",
            "\u001b[0m"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kD9DO1Is9Edf",
        "colab_type": "code",
        "outputId": "ab502fe8-3cce-46f6-ad5d-31d1de80f603",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1892
        }
      },
      "source": [
        "#Run the training function. \n",
        "#Output is the data dimension after each Convo & Dropout  operation has been applied. \n",
        "!pipenv run python train.py --epochs 2 --save_frequency 2 /content/data/cars/steelroses/collected #path to data"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[31m\u001b[1mWarning\u001b[39m\u001b[22m: Your Pipfile requires \u001b[39m\u001b[1mpython_version\u001b[39m\u001b[22m \u001b[34m\u001b[22m3.*\u001b[39m\u001b[22m, but you are using \u001b[34m\u001b[22m3.6.7\u001b[39m\u001b[22m (\u001b[32m\u001b[22m/root/.local/share/v/f/bin/python\u001b[39m\u001b[22m).\n",
            "  \u001b[32m\u001b[22m$ pipenv --rm\u001b[39m\u001b[22m and rebuilding the virtual environment may resolve the issue.\n",
            "  \u001b[31m\u001b[22m$ pipenv check\u001b[39m\u001b[22m will surely fail.\n",
            "Using TensorFlow backend.\n",
            "tcmalloc: large alloc 6994501632 bytes == 0x4b50000 @  0x7f38b56f7001 0x7f38b313ede5 0x7f38b31a36f1 0x7f38b31a57cf 0x7f38b323e158 0x5030d5 0x506859 0x504c28 0x506393 0x634d52 0x634e0a 0x6385c8 0x63915a 0x4a6f10 0x7f38b52f2b97 0x5afa0a\n",
            "tcmalloc: large alloc 3497254912 bytes == 0x1a59ca000 @  0x7f38b56f51e7 0x7f38b313eca1 0x7f38b31a6580 0x7f38b3236a82 0x5030d5 0x506859 0x504c28 0x506393 0x634d52 0x634e0a 0x6385c8 0x63915a 0x4a6f10 0x7f38b52f2b97 0x5afa0a\n",
            "adding first convolutional layer\n",
            "WARNING:tensorflow:From /root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Colocations handled automatically by placer.\n",
            "WARNING:tensorflow:From /root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:3445: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n",
            "adding second convolutional layer\n",
            "adding third convolutional layer\n",
            "adding fourth convolutional layer\n",
            "adding fifth convolutional layer\n",
            "adding fully connected layer\n",
            "adding output layer\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d_1 (Conv2D)            (None, 18, 64, 24)        1824      \n",
            "_________________________________________________________________\n",
            "dropout_1 (Dropout)          (None, 18, 64, 24)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_2 (Conv2D)            (None, 9, 32, 32)         19232     \n",
            "_________________________________________________________________\n",
            "dropout_2 (Dropout)          (None, 9, 32, 32)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_3 (Conv2D)            (None, 5, 16, 40)         32040     \n",
            "_________________________________________________________________\n",
            "dropout_3 (Dropout)          (None, 5, 16, 40)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_4 (Conv2D)            (None, 3, 8, 48)          17328     \n",
            "_________________________________________________________________\n",
            "dropout_4 (Dropout)          (None, 3, 8, 48)          0         \n",
            "_________________________________________________________________\n",
            "conv2d_5 (Conv2D)            (None, 2, 4, 48)          20784     \n",
            "_________________________________________________________________\n",
            "dropout_5 (Dropout)          (None, 2, 4, 48)          0         \n",
            "_________________________________________________________________\n",
            "flatten_1 (Flatten)          (None, 384)               0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 100)               38500     \n",
            "_________________________________________________________________\n",
            "dropout_6 (Dropout)          (None, 100)               0         \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 129,809\n",
            "Trainable params: 129,809\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n",
            "starting epoch 0\n",
            "WARNING:tensorflow:From /root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use tf.cast instead.\n",
            "Train on 56921 samples, validate on 6325 samples\n",
            "Epoch 1/1\n",
            "2019-06-11 22:37:23.721323: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA\n",
            "2019-06-11 22:37:23.827073: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2300000000 Hz\n",
            "2019-06-11 22:37:23.831546: I tensorflow/compiler/xla/service/service.cc:150] XLA service 0x261a840 executing computations on platform Host. Devices:\n",
            "2019-06-11 22:37:23.831589: I tensorflow/compiler/xla/service/service.cc:158]   StreamExecutor device (0): <undefined>, <undefined>\n",
            "2019-06-11 22:37:24.715417: W tensorflow/core/framework/allocator.cc:124] Allocation of 20480000 exceeds 10% of system memory.\n",
            "2019-06-11 22:37:24.715799: W tensorflow/core/framework/allocator.cc:124] Allocation of 20480000 exceeds 10% of system memory.\n",
            "2019-06-11 22:37:24.780844: W tensorflow/core/framework/allocator.cc:124] Allocation of 27648000 exceeds 10% of system memory.\n",
            "2019-06-11 22:37:24.814182: W tensorflow/core/framework/allocator.cc:124] Allocation of 23500800 exceeds 10% of system memory.\n",
            "   25/56921 [..............................] - ETA: 43:32 - loss: 0.9303 - mean_squared_error: 0.93032019-06-11 22:37:25.144582: W tensorflow/core/framework/allocator.cc:124] Allocation of 20480000 exceeds 10% of system memory.\n",
            "56921/56921 [==============================] - 934s 16ms/step - loss: 0.6801 - mean_squared_error: 0.6801 - val_loss: 2.6742 - val_mean_squared_error: 2.6742\n",
            "Saving epoch 0 to weights_2019-06-11_22-36-47_epoch_0.h5\n",
            "starting epoch 1\n",
            "Train on 56921 samples, validate on 6325 samples\n",
            "Epoch 1/1\n",
            "56921/56921 [==============================] - 878s 15ms/step - loss: 0.5711 - mean_squared_error: 0.5711 - val_loss: 2.9971 - val_mean_squared_error: 2.9971\n",
            "starting epoch 2\n",
            "Train on 56921 samples, validate on 6325 samples\n",
            "Epoch 1/1\n",
            "56921/56921 [==============================] - 821s 14ms/step - loss: 0.5109 - mean_squared_error: 0.5109 - val_loss: 3.2398 - val_mean_squared_error: 3.2398\n",
            "Saving epoch 2 to weights_2019-06-11_22-36-47_epoch_2.h5\n",
            "starting epoch 3\n",
            "Train on 56921 samples, validate on 6325 samples\n",
            "Epoch 1/1\n",
            "56921/56921 [==============================] - 839s 15ms/step - loss: 0.4612 - mean_squared_error: 0.4612 - val_loss: 2.7540 - val_mean_squared_error: 2.7540\n",
            "starting epoch 4\n",
            "Train on 56921 samples, validate on 6325 samples\n",
            "Epoch 1/1\n",
            "56921/56921 [==============================] - 876s 15ms/step - loss: 0.4177 - mean_squared_error: 0.4177 - val_loss: 3.0711 - val_mean_squared_error: 3.0711\n",
            "Saving epoch 4 to weights_2019-06-11_22-36-47_epoch_4.h5\n",
            "starting epoch 5\n",
            "Train on 56921 samples, validate on 6325 samples\n",
            "Epoch 1/1\n",
            " 2375/56921 [>.............................] - ETA: 14:53 - loss: 0.3959 - mean_squared_error: 0.3959Traceback (most recent call last):\n",
            "  File \"train.py\", line 96, in <module>\n",
            "    batch_size=25, epochs=1, verbose=1, validation_split=0.1, shuffle=True)\n",
            "  File \"/root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/keras/engine/training.py\", line 1039, in fit\n",
            "    validation_steps=validation_steps)\n",
            "  File \"/root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/keras/engine/training_arrays.py\", line 199, in fit_loop\n",
            "    outs = f(ins_batch)\n",
            "  File \"/root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py\", line 2715, in __call__\n",
            "    return self._call(inputs)\n",
            "  File \"/root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py\", line 2675, in _call\n",
            "    fetched = self._callable_fn(*array_vals)\n",
            "  File \"/root/.local/share/virtualenvs/foocars--MKvUarz/lib/python3.6/site-packages/tensorflow/python/client/session.py\", line 1439, in __call__\n",
            "    run_metadata_ptr)\n",
            "KeyboardInterrupt\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "49Un0mi-x8RZ",
        "colab_type": "code",
        "outputId": "dc741059-426b-448f-fc4c-9d178e5d8146",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 173
        }
      },
      "source": [
        "!pip list | grep tensor #check Tensorflow is installed\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "mesh-tensorflow          0.0.5                \n",
            "tensor2tensor            1.11.0               \n",
            "tensorboard              1.13.1               \n",
            "tensorboardcolab         0.0.22               \n",
            "tensorflow               1.13.1               \n",
            "tensorflow-estimator     1.13.0               \n",
            "tensorflow-hub           0.4.0                \n",
            "tensorflow-metadata      0.13.0               \n",
            "tensorflow-probability   0.6.0                \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "InYmY9lkxmrc",
        "colab_type": "text"
      },
      "source": [
        "## What are steerstats \n",
        "\n",
        "Let's look"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fO7U8sqpxrd-",
        "colab_type": "code",
        "outputId": "3f37dab3-2ba7-4dc5-b785-b8875135a1df",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        }
      },
      "source": [
        "#We can see the \"weights....\" files\n",
        "!ls"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "defines.py\t  readme.md\t    weights_2019-06-11_22-36-47_epoch_0.h5\n",
            "dropout_model.py  steerstats.npz    weights_2019-06-11_22-36-47_epoch_2.h5\n",
            "foocars\t\t  train_history.py  weights_2019-06-11_22-36-47_epoch_4.h5\n",
            "history_model.py  train.py\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2vVZnbXmxs0j",
        "colab_type": "text"
      },
      "source": [
        "## Download weights and steerstats"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZAwcWUHr7-uw",
        "colab_type": "code",
        "outputId": "6528c09e-70ed-4d05-d9f9-2b933d7c586d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "!pip install tqdm #gadget to show progress bar\n",
        "from tqdm import tqdm"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: tqdm in /usr/local/lib/python3.6/dist-packages (4.28.1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z0GBQCFNyOOT",
        "colab_type": "code",
        "outputId": "563e83c2-7d4c-4870-8080-473aeb3c42d9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        }
      },
      "source": [
        "# get car files and download\n",
        "!tar cvzf car.weights.tar.gz *.h5 *.npz\n",
        "!rm *.h5 *.npz"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "weights_2019-06-11_22-36-47_epoch_0.h5\n",
            "weights_2019-06-11_22-36-47_epoch_2.h5\n",
            "weights_2019-06-11_22-36-47_epoch_4.h5\n",
            "steerstats.npz\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Kuf8QyfuzGb9",
        "colab_type": "code",
        "outputId": "4329161d-1361-4720-9ffc-f5e8158c1c29",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "#remove results (to cleanup)\n",
        "!rm *.h5 *.npz"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "rm: cannot remove '*.h5': No such file or directory\n",
            "rm: cannot remove '*.npz': No such file or directory\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fs0qI4NjzNJ4",
        "colab_type": "code",
        "outputId": "24c1127e-6200-4e42-c810-5475764bfb06",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        }
      },
      "source": [
        "!mkdir -p /content/data/cars/steelroses/model\n",
        "!tar xvzf car.weights.tar.gz -C /content/data/cars/steelroses/model/"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "weights_2019-06-11_22-36-47_epoch_0.h5\n",
            "weights_2019-06-11_22-36-47_epoch_2.h5\n",
            "weights_2019-06-11_22-36-47_epoch_4.h5\n",
            "steerstats.npz\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4By3H84LymGN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Download the weights files containing the output\n",
        "files.download('car.weights.tar.gz')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qT6hRvOMLgPq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BdEHNyjI0CWk",
        "colab_type": "text"
      },
      "source": [
        "## Examine the weights and steer stats\n",
        "\n",
        "What does the computer think should happen?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RwiGnlcR0KOE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "steerstats = np.load('/content/data/cars/steelroses/model/steerstats.npz')\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MZLhj4v40xKT",
        "colab_type": "code",
        "outputId": "5814c70f-d54a-4d1c-b447-6afe11801018",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "print(type(steerstats))\n",
        "steerstats.keys()\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<class 'numpy.lib.npyio.NpzFile'>\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "KeysView(<numpy.lib.npyio.NpzFile object at 0x7fdae1f45128>)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WJ6E_aUI1UVP",
        "colab_type": "code",
        "outputId": "c6965264-c179-4ab7-dac9-ee95ea7157a4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "steerstats['arr_0'][0:10]\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([1394.95028935,  146.93441373])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VKHhp7de7SJJ",
        "colab_type": "text"
      },
      "source": [
        "## Examine the collected data\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V2jjkl1f49Ga",
        "colab_type": "code",
        "outputId": "2c6e58f7-2e71-43c6-e89a-dd5100562a43",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# commands.npz\n",
        "filename = '/content/data/cars/steelroses/collected/commands_2018-12-06_12-54-17.npz'\n",
        "commands = np.load( filename )\n",
        "print(type(commands))\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<class 'numpy.lib.npyio.NpzFile'>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S4b5nQBX5Cqe",
        "colab_type": "code",
        "outputId": "f01549dd-7cf7-43d3-c70c-91a345424f3a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 208
        }
      },
      "source": [
        "print(commands.keys())\n",
        "commands['arr_0'][0:10]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "KeysView(<numpy.lib.npyio.NpzFile object at 0x7fdae1f45390>)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[1411., 1502.],\n",
              "       [1411., 1501.],\n",
              "       [1411., 1501.],\n",
              "       [1412., 1501.],\n",
              "       [1412., 1501.],\n",
              "       [1411., 1502.],\n",
              "       [1411., 1501.],\n",
              "       [1411., 1502.],\n",
              "       [1412., 1501.],\n",
              "       [1412., 1501.]], dtype=float16)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 29
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xgx1KXBF5Cxa",
        "colab_type": "code",
        "outputId": "57f2281e-87a8-4b50-aa9f-786b05db95c3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# IME npz\n",
        "imu = np.load('/content/data/cars/steelroses/collected/IMU_2018-12-06_12-54-17.npz')\n",
        "print(type(imu))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<class 'numpy.lib.npyio.NpzFile'>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hzXYqEaZ4F1m",
        "colab_type": "code",
        "outputId": "4e353185-acfc-491a-f67d-79d5e0e47f4e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 381
        }
      },
      "source": [
        "print(imu.keys())\n",
        "imu['arr_0'][0:10]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "KeysView(<numpy.lib.npyio.NpzFile object at 0x7fdae34948d0>)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2777988.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926810.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926840.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926850.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926880.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926910.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926930.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926950.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2926990.],\n",
              "       [      0.,       0.,       0.,       0.,       0.,       0.,\n",
              "        2927000.]], dtype=float32)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 31
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JqU42eNG4axC",
        "colab_type": "text"
      },
      "source": [
        "## What images were collected"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t-t3B0jt4dQP",
        "colab_type": "code",
        "outputId": "e5db3151-140a-4f4f-956b-b63edece895e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "images = np.load('/content/data/cars/steelroses/collected/imgs_2018-12-06_12-54-17.npz')\n",
        "print(type(images))\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<class 'numpy.lib.npyio.NpzFile'>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GqAWBP0n4q_z",
        "colab_type": "code",
        "outputId": "2868c98b-d1d9-4940-94e1-eae74c1e1033",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 884
        }
      },
      "source": [
        "print(images.keys())\n",
        "images['arr_0'][0]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "KeysView(<numpy.lib.npyio.NpzFile object at 0x7fdae1f4ce80>)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[[ 32,  50,  78],\n",
              "        [113, 130, 159],\n",
              "        [170, 198, 223],\n",
              "        ...,\n",
              "        [ 12,   0,   0],\n",
              "        [ 11,   0,   0],\n",
              "        [ 32,  15,   5]],\n",
              "\n",
              "       [[ 46,  64,  92],\n",
              "        [137, 155, 183],\n",
              "        [158, 185, 210],\n",
              "        ...,\n",
              "        [ 13,   0,   0],\n",
              "        [ 11,   0,   0],\n",
              "        [ 32,  15,   5]],\n",
              "\n",
              "       [[100, 124, 153],\n",
              "        [170, 194, 223],\n",
              "        [151, 170, 191],\n",
              "        ...,\n",
              "        [ 15,   0,   0],\n",
              "        [ 12,   0,   0],\n",
              "        [ 33,  16,   6]],\n",
              "\n",
              "       ...,\n",
              "\n",
              "       [[115, 118,  94],\n",
              "        [114, 117,  93],\n",
              "        [107, 110,  86],\n",
              "        ...,\n",
              "        [249, 228, 186],\n",
              "        [255, 235, 192],\n",
              "        [247, 226, 183]],\n",
              "\n",
              "       [[116, 120,  94],\n",
              "        [108, 112,  85],\n",
              "        [104, 108,  82],\n",
              "        ...,\n",
              "        [255, 235, 192],\n",
              "        [254, 233, 192],\n",
              "        [244, 223, 183]],\n",
              "\n",
              "       [[105, 109,  83],\n",
              "        [107, 110,  84],\n",
              "        [108, 112,  85],\n",
              "        ...,\n",
              "        [255, 234, 191],\n",
              "        [249, 228, 188],\n",
              "        [232, 210, 170]]], dtype=uint8)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 33
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l4qWGFfM8Aml",
        "colab_type": "code",
        "outputId": "57bf0998-b13f-4aa9-c5e6-4dd1c667e35b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 269
        }
      },
      "source": [
        "\n",
        "imgplot = plt.imshow(images['arr_0'][0])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUcAAAD8CAYAAADkM2ZpAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRYs2wuKXY7Ro0aKNsP8HkXV/kKDXD6IAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JG77ByEIJGrL",
        "colab_type": "text"
      },
      "source": [
        "**NOTE**: change the location of IMG_IN accordingly  "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EBfxHmSS-0e7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "## Make a video clip\n",
        "# Tutorial: https://www.apriorit.com/dev-blog/600-colab-for-video-processing\n",
        "\n",
        "import cv2\n",
        "VIDEO_FORMAT = 'VP90'\n",
        "VIDEO_EXT = 'webm'\n",
        "\n",
        "!mkdir -p /content/data/output\n",
        "\n",
        "VIDEO_OUT = '/content/data/output.' + VIDEO_EXT\n",
        "#IMG_IN = '/content/data/cars/steelroses/collected/imgs_2018-12-06_12-54-17.npz'\n",
        "IMG_IN    = '/content/collected/imgs_2018-12-06_12-54-17.npz' #CHANGE THIS ACCORDINGLY"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZTM4FmEE__f6",
        "colab_type": "code",
        "outputId": "14d79e08-ae8b-45fa-d9bb-f095a774e8e5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "#THIS IS A TEST\n",
        "image = images['arr_0'][0]\n",
        "print(image.shape)\n",
        "\n",
        "fourcc = cv2.VideoWriter_fourcc(*VIDEO_FORMAT)\n",
        "writer = cv2.VideoWriter(VIDEO_OUT, fourcc, 30, (image.shape[1], image.shape[0]), True)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(96, 128, 3)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M-EyQB0iIe0L",
        "colab_type": "code",
        "outputId": "9e897bc6-b899-4dd3-a38b-6f9eeb5ddd18",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 377
        }
      },
      "source": [
        "\n",
        "imgdata = np.load(IMG_IN)['arr_0']\n",
        "print(f'\\nShape: {imgdata.shape}\\n')\n",
        "\n",
        "\n",
        "#ITERATE  the first dimension and get the image from the rest\n",
        "for index in tqdm(range(0,len(imgdata))):\n",
        "  img = imgdata[index]\n",
        "  writer.write(img)\n",
        "\n",
        "writer.release()\n",
        "print(f'Done: {VIDEO_OUT}')\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "error",
          "ename": "FileNotFoundError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mFileNotFoundError\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-37-3985c84a295e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mimgdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mIMG_IN\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'arr_0'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34mf'\\nShape: {imgdata.shape}\\n'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/numpy/lib/npyio.py\u001b[0m in \u001b[0;36mload\u001b[0;34m(file, mmap_mode, allow_pickle, fix_imports, encoding)\u001b[0m\n\u001b[1;32m    420\u001b[0m         \u001b[0mown_fid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    421\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 422\u001b[0;31m         \u001b[0mfid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mopen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos_fspath\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfile\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"rb\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    423\u001b[0m         \u001b[0mown_fid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    424\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mFileNotFoundError\u001b[0m: [Errno 2] No such file or directory: '/content/collected/imgs_2018-12-06_12-54-17.npz'"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SbYspbgSWXTm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "!ls /content/data"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uVkn-ZcctBnP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Save video locally\n",
        "#will save to: C:\\Users\\<username>\\Downloads\n",
        "files.download(VIDEO_OUT)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XuwUvkWLcrmK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from IPython.core.display import Video\n",
        "\n",
        "Video(VIDEO_OUT)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RTPyX9i8YTVx",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#THIS IS A TEST\n",
        "# The dynamic video tag is not showing. Maybe, do this in javacsript\n",
        "from IPython.core.display import display, HTML\n",
        "\n",
        "\n",
        "HTML_SRC = \"\"\"<video width=\"400\" height=\"200\">\n",
        "<source src=\"/content/data/output.webm\" type=\"video/x-webm\"></source>\n",
        "</video>\"\"\"\n",
        "print(HTML_SRC)\n",
        "\n",
        "\n",
        "\n",
        "             \n",
        "display(HTML(HTML_SRC))             "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p3jiLd13thvA",
        "colab_type": "text"
      },
      "source": [
        "# Consulting the Model\n",
        "Uploading the trained weights\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BMoFo8XN3sO-",
        "colab_type": "code",
        "outputId": "2f64f952-d65c-448f-9ac0-c658ac4a54b6",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 904
        }
      },
      "source": [
        "import keras\n",
        "import tensorflow as tf\n",
        "import concurrent.futures\n",
        "from dropout_model import model\n",
        "from defines import *\n",
        "\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "adding first convolutional layer\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Colocations handled automatically by placer.\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3445: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n",
            "adding second convolutional layer\n",
            "adding third convolutional layer\n",
            "adding fourth convolutional layer\n",
            "adding fifth convolutional layer\n",
            "adding fully connected layer\n",
            "adding output layer\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d_1 (Conv2D)            (None, 18, 64, 24)        1824      \n",
            "_________________________________________________________________\n",
            "dropout_1 (Dropout)          (None, 18, 64, 24)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_2 (Conv2D)            (None, 9, 32, 32)         19232     \n",
            "_________________________________________________________________\n",
            "dropout_2 (Dropout)          (None, 9, 32, 32)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_3 (Conv2D)            (None, 5, 16, 40)         32040     \n",
            "_________________________________________________________________\n",
            "dropout_3 (Dropout)          (None, 5, 16, 40)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_4 (Conv2D)            (None, 3, 8, 48)          17328     \n",
            "_________________________________________________________________\n",
            "dropout_4 (Dropout)          (None, 3, 8, 48)          0         \n",
            "_________________________________________________________________\n",
            "conv2d_5 (Conv2D)            (None, 2, 4, 48)          20784     \n",
            "_________________________________________________________________\n",
            "dropout_5 (Dropout)          (None, 2, 4, 48)          0         \n",
            "_________________________________________________________________\n",
            "flatten_1 (Flatten)          (None, 384)               0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 100)               38500     \n",
            "_________________________________________________________________\n",
            "dropout_6 (Dropout)          (None, 100)               0         \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 129,809\n",
            "Trainable params: 129,809\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c2hRE_HVwSkj",
        "colab_type": "code",
        "outputId": "1136cd48-6e22-4865-dbeb-99426199d0bf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "! ls /content/data/cars/steelroses/model"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "steerstats.npz\t\t\t\tweights_2019-06-11_22-36-47_epoch_2.h5\n",
            "weights_2019-06-11_22-36-47_epoch_0.h5\tweights_2019-06-11_22-36-47_epoch_4.h5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FreGpfk_vvas",
        "colab_type": "code",
        "outputId": "0aaf7c09-beca-4399-8ea9-fa377adae561",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "DATA_DIR = '/content/data/cars'\n",
        "CAR_NAME = 'steelroses'\n",
        "WEIGHTS_DIR = DATA_DIR + \"/\" + CAR_NAME + '/model/'\n",
        "WEIGHTS_FILE = WEIGHTS_DIR + 'weights_2019-04-06_13-00-23_epoch_4.h5'\n",
        "STEERSTATS_FILE = DATA_DIR + \"/\" + CAR_NAME + '/model/steerstats.npz'\n",
        "print(WEIGHTS_FILE)\n",
        "print(STEERSTATS_FILE)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/data/cars/steelroses/model/weights_2019-04-06_13-00-23_epoch_4.h5\n",
            "/content/data/cars/steelroses/model/steerstats.npz\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y01QwVTmxdTF",
        "colab_type": "code",
        "outputId": "17e9376c-557f-4d03-9966-339b9158eed2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 763
        }
      },
      "source": [
        "# We upload( put model in memory) the trained model to analyze it\n",
        "#same model as train.py \n",
        "\n",
        "# Use Dropout Model\n",
        "# https://github.com/fubarlabs/foocars/blob/1e6b2bccac5b8595fd053816d9117169d35ae3a1/training/dropout_model.py\n",
        "\n",
        "\n",
        "\n",
        "import keras\n",
        "from keras.models import Sequential\n",
        "from keras.layers.core import Dense, Dropout, Activation, Flatten, Reshape\n",
        "from keras.layers import Embedding, Input, merge\n",
        "from keras.layers.convolutional import Conv2D, MaxPooling2D\n",
        "from keras.optimizers import Adam, SGD\n",
        "from keras.regularizers import l2, l1\n",
        "from keras.utils.np_utils import to_categorical\n",
        "from keras.layers.normalization import BatchNormalization\n",
        "from keras import backend as K\n",
        "\n",
        "\n",
        "\n",
        "nrows=36\n",
        "ncols=128\n",
        "wr=0.00001 # l1 regularizer value\n",
        "dp=0.125 # dropout rate \n",
        "\n",
        "# Note: Dan used the keras functional paradigm to define his network.\n",
        "# I'm using the sequential paradigm. \n",
        "model=Sequential()\n",
        "frame_in = Input(shape=(3, nrows, ncols), name='img_input')\n",
        "\n",
        "#we should do a local contrast normalization\n",
        "\n",
        "print(\"adding first convolutional layer\")\n",
        "#5x5 convolutional layer with a stride of 2\n",
        "#model.add(BatchNormalization(input_shape=(nrows, ncols, 3)))\n",
        "model.add(Conv2D(24, (5, 5), input_shape=(nrows, ncols, 3), strides=(2, 2), activation='elu', padding='same', kernel_initializer='lecun_uniform'))\n",
        "#model.add(MaxPooling2D(pool_size=(2, 2), data_format=\"channels_last\"))\n",
        "model.add(Dropout(dp))\n",
        "\n",
        "print(\"adding second convolutional layer\")\n",
        "#5x5 convolutional layer with a stride of 2\n",
        "#model.add(BatchNormalization())\n",
        "model.add(Conv2D(32, (5, 5), strides=(2, 2), activation='elu', padding='same', kernel_initializer='lecun_uniform'))\n",
        "#model.add(MaxPooling2D(pool_size=(2, 2), data_format=\"channels_last\"))\n",
        "model.add(Dropout(dp))\n",
        "\n",
        "print(\"adding third convolutional layer\")\n",
        "#5x5 convolutional layer with a stride of 2\n",
        "#model.add(BatchNormalization())\n",
        "model.add(Conv2D(40, (5, 5), strides=(2, 2), activation='elu', padding='same', kernel_initializer='lecun_uniform'))\n",
        "#model.add(MaxPooling2D(pool_size=(2, 2), data_format=\"channels_last\"))\n",
        "model.add(Dropout(dp))\n",
        "\n",
        "print(\"adding fourth convolutional layer\")\n",
        "#3x3 convolutional layer with no stride \n",
        "#model.add(BatchNormalization())\n",
        "model.add(Conv2D(48, (3, 3), strides=(2, 2), activation='elu', padding='same', kernel_initializer='lecun_uniform'))\n",
        "#model.add(MaxPooling2D(pool_size=(2, 2), data_format=\"channels_last\"))\n",
        "model.add(Dropout(dp))\n",
        "\n",
        "print(\"adding fifth convolutional layer\")\n",
        "#3x3 convolutional layer with no stride \n",
        "#model.add(BatchNormalization())\n",
        "model.add(Conv2D(48, (3, 3), strides=(2, 2), activation='elu', padding='same', kernel_initializer='lecun_uniform'))\n",
        "#model.add(MaxPooling2D(pool_size=(2, 2), data_format=\"channels_last\"))\n",
        "#model.add(BatchNormalization())\n",
        "model.add(Dropout(dp))\n",
        "\n",
        "\n",
        "model.add(Flatten())\n",
        "\n",
        "print(\"adding fully connected layer\")\n",
        "#fully connected layer\n",
        "model.add(Dense(100, activation='elu', kernel_initializer='lecun_uniform'))\n",
        "model.add(Dropout(dp))\n",
        "\n",
        "print(\"adding output layer\")\n",
        "#fully connected layer to output node\n",
        "model.add(Dense(1, activation='linear', kernel_initializer='lecun_uniform'))\n",
        "\n",
        "model.compile(loss=['mse'], optimizer=SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=True), metrics=['mse'])\n",
        "print(model.summary())\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "adding first convolutional layer\n",
            "adding second convolutional layer\n",
            "adding third convolutional layer\n",
            "adding fourth convolutional layer\n",
            "adding fifth convolutional layer\n",
            "adding fully connected layer\n",
            "adding output layer\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d_6 (Conv2D)            (None, 18, 64, 24)        1824      \n",
            "_________________________________________________________________\n",
            "dropout_7 (Dropout)          (None, 18, 64, 24)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_7 (Conv2D)            (None, 9, 32, 32)         19232     \n",
            "_________________________________________________________________\n",
            "dropout_8 (Dropout)          (None, 9, 32, 32)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_8 (Conv2D)            (None, 5, 16, 40)         32040     \n",
            "_________________________________________________________________\n",
            "dropout_9 (Dropout)          (None, 5, 16, 40)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_9 (Conv2D)            (None, 3, 8, 48)          17328     \n",
            "_________________________________________________________________\n",
            "dropout_10 (Dropout)         (None, 3, 8, 48)          0         \n",
            "_________________________________________________________________\n",
            "conv2d_10 (Conv2D)           (None, 2, 4, 48)          20784     \n",
            "_________________________________________________________________\n",
            "dropout_11 (Dropout)         (None, 2, 4, 48)          0         \n",
            "_________________________________________________________________\n",
            "flatten_2 (Flatten)          (None, 384)               0         \n",
            "_________________________________________________________________\n",
            "dense_3 (Dense)              (None, 100)               38500     \n",
            "_________________________________________________________________\n",
            "dropout_12 (Dropout)         (None, 100)               0         \n",
            "_________________________________________________________________\n",
            "dense_4 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 129,809\n",
            "Trainable params: 129,809\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PKfqNzEcuk_z",
        "colab_type": "code",
        "outputId": "daf74e7d-9d0f-4248-c4d4-4a103fcbccd5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 505
        }
      },
      "source": [
        " # Take the trained weights file and load into the model\n",
        "  model.load_weights(WEIGHTS_FILE)\n",
        "  model._make_predict_function()\n",
        "  global g_steerstats #steering info\n",
        "  g_steerstats=np.load(STEERSTATS_FILE)['arr_0']\n",
        "  \n",
        "  #g_camera=picamera.PiCamera()\n",
        "  #g_camera.resolution=(128, 96)\n",
        "  #g_camera.framerate=FRAME_RATE"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "error",
          "ename": "OSError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-42-546e04dd5b13>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload_weights\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mWEIGHTS_FILE\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_make_predict_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;32mglobal\u001b[0m \u001b[0mg_steerstats\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0mg_steerstats\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mSTEERSTATS_FILE\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'arr_0'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/keras/engine/network.py\u001b[0m in \u001b[0;36mload_weights\u001b[0;34m(self, filepath, by_name, skip_mismatch, reshape)\u001b[0m\n\u001b[1;32m   1155\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mh5py\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1156\u001b[0m             \u001b[0;32mraise\u001b[0m \u001b[0mImportError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'`load_weights` requires h5py.'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1157\u001b[0;31m         \u001b[0;32mwith\u001b[0m \u001b[0mh5py\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mFile\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfilepath\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmode\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'r'\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1158\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0;34m'layer_names'\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mattrs\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0;34m'model_weights'\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1159\u001b[0m                 \u001b[0mf\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'model_weights'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/h5py/_hl/files.py\u001b[0m in \u001b[0;36m__init__\u001b[0;34m(self, name, mode, driver, libver, userblock_size, swmr, **kwds)\u001b[0m\n\u001b[1;32m    310\u001b[0m             \u001b[0;32mwith\u001b[0m \u001b[0mphil\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    311\u001b[0m                 \u001b[0mfapl\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmake_fapl\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdriver\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlibver\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwds\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 312\u001b[0;31m                 \u001b[0mfid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmake_fid\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmode\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0muserblock_size\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfapl\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mswmr\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mswmr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    313\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    314\u001b[0m                 \u001b[0;32mif\u001b[0m \u001b[0mswmr_support\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/h5py/_hl/files.py\u001b[0m in \u001b[0;36mmake_fid\u001b[0;34m(name, mode, userblock_size, fapl, fcpl, swmr)\u001b[0m\n\u001b[1;32m    140\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mswmr\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mswmr_support\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    141\u001b[0m             \u001b[0mflags\u001b[0m \u001b[0;34m|=\u001b[0m \u001b[0mh5f\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mACC_SWMR_READ\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 142\u001b[0;31m         \u001b[0mfid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mh5f\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mopen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mflags\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfapl\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mfapl\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    143\u001b[0m     \u001b[0;32melif\u001b[0m \u001b[0mmode\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;34m'r+'\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    144\u001b[0m         \u001b[0mfid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mh5f\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mopen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mh5f\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mACC_RDWR\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfapl\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mfapl\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32mh5py/_objects.pyx\u001b[0m in \u001b[0;36mh5py._objects.with_phil.wrapper\u001b[0;34m()\u001b[0m\n",
            "\u001b[0;32mh5py/_objects.pyx\u001b[0m in \u001b[0;36mh5py._objects.with_phil.wrapper\u001b[0;34m()\u001b[0m\n",
            "\u001b[0;32mh5py/h5f.pyx\u001b[0m in \u001b[0;36mh5py.h5f.open\u001b[0;34m()\u001b[0m\n",
            "\u001b[0;31mOSError\u001b[0m: Unable to open file (unable to open file: name = '/content/data/cars/steelroses/model/weights_2019-04-06_13-00-23_epoch_4.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xqAQZOVQyseX",
        "colab_type": "code",
        "outputId": "f834645f-2595-43c8-f7a4-9dd047a512da",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 377
        }
      },
      "source": [
        "#defines image size:\n",
        "row_offset=30\n",
        "nrows=36\n",
        "ncols=128\n",
        "\n",
        "\n",
        "#imgdata = np.load(IMG_IN)['arr_0'].astype('float32')\n",
        "imgdata = np.load(IMG_IN)['arr_0'] #give it an image\n",
        "print(f'\\nShape: {imgdata.shape}\\n')\n",
        "\n",
        "index = 1\n",
        "img = imgdata[index]\n",
        "\n",
        "# Image must be in the shape the model is using\n",
        "#ValueError: Error when checking input: expected conv2d_6_input to have shape (36, 128, 3) but got array with shape (96, 128, 3)\n",
        "# original image shape (200, 96, 128, 3)\n",
        "\n",
        "#imagerawdata=np.reshape(np.fromstring(s, dtype=np.uint8), (96, 128, 3), 'C')\n",
        "#imdata=imagerawdata[20:56, :]\n",
        "\n",
        "#pre process - cleaning and standardizing\n",
        "immean=img.mean()\n",
        "imvar=img.std()\n",
        "g_imageData=np.copy((imgdata-immean)/imvar) #standardizing \n",
        "fimg = g_imageData[0]\n",
        "\n",
        "crop_image=fimg[row_offset:row_offset+nrows, :]\n",
        "image_mean=crop_image.mean()\n",
        "image_std=crop_image.std()\n",
        "guess_image=(crop_image-image_mean)/image_std\n",
        "\n",
        "\n",
        "#The actual prediction (!!!!)\n",
        "#I give you an image, you give me back the steering decision !!!\n",
        "pred=model.predict(np.expand_dims(guess_image, axis=0))      \n",
        "steer_command=pred[0][0]*g_steerstats[1]+g_steerstats[0]\n",
        "\n",
        "print(f'pred: {pred}, steer: {steer_command}')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "error",
          "ename": "FileNotFoundError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mFileNotFoundError\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-43-56cdce685f9e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0;31m#imgdata = np.load(IMG_IN)['arr_0'].astype('float32')\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m \u001b[0mimgdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mIMG_IN\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'arr_0'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      8\u001b[0m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34mf'\\nShape: {imgdata.shape}\\n'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      9\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/numpy/lib/npyio.py\u001b[0m in \u001b[0;36mload\u001b[0;34m(file, mmap_mode, allow_pickle, fix_imports, encoding)\u001b[0m\n\u001b[1;32m    420\u001b[0m         \u001b[0mown_fid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    421\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 422\u001b[0;31m         \u001b[0mfid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mopen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos_fspath\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfile\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"rb\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    423\u001b[0m         \u001b[0mown_fid\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    424\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mFileNotFoundError\u001b[0m: [Errno 2] No such file or directory: '/content/collected/imgs_2018-12-06_12-54-17.npz'"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fBGeHrKAciWP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!ls"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}