Intent Recognition using pytorch

PyTorch is a deep learning framework that puts Python first. This project is an intent recognisor using pytorch. Intent recognition is a natural language processing task for finding what are the actions specfied in a sentence.

Used in building virtual assistants

Get the code

Run this command in terminal / command prompt

git clone https://github.com/GopikrishnanSasikumar/Text_Classifier-pytorch.git

You need:

• python >= 3.0

  Install python3

• pip

  For linux:

  sudo apt-get install python3-pip
  

  For Mac

  To install or upgrade pip, download get-pip.py from the official site. Then run the following command:

  sudo python get-pip.py
  

• pytorch

  For mac:

  pip install http://download.pytorch.org/whl/torch-0.1.12.post2-cp35-cp35m-macosx_10_7_x86_64.whl
  

  pip install torchvision
  

  For Linux

  pip install http://download.pytorch.org/whl/cu75/torch-0.1.12.post2-cp35-cp35m-linux_x86_64.whl
  

  pip install torchvision
  

Running

Run

python3 train.py

for training the neural network model it Will create and store the model in ann.pt

Testing

python3 test.py

will initiate an interactive section like this

User:

Enter the sentence and see the output.

What's going on ?

A neural network in pytorch can be implemented like this.

class ANN(nn.Module):

    def __init__(self,input_size,hidden_size,output_size):
        super(ANN,self).__init__()
        self.i2h = nn.Linear(input_size,hidden_size) #input to hidden layer
        self.h2o = nn.Linear(hidden_size,output_size) #hidden to output layer
        self.softmax = nn.LogSoftmax() #softmax layer

    def forward(self, input):
        # forward pass of the network
        hidden = self.i2h(input)
        output = self.h2o(hidden)
        output = self.softmax(output)
        return output

Dataset used for training the neural network is initialized in a list like this,

training_data = list()
training_data.append({"intent":"greeting", "sentence":"how are you?"})
training_data.append({"intent":"greeting", "sentence":"how is your day?"})
training_data.append({"intent":"greeting", "sentence":"hi there hello"})
training_data.append({"intent":"greeting", "sentence":"good morning"})
training_data.append({"intent":"greeting", "sentence":"good day"})
training_data.append({"intent":"greeting", "sentence":"how is it going today?"})

training_data.append({"intent":"goodbye", "sentence":"have a nice day"})
training_data.append({"intent":"goodbye", "sentence":"see you later"})
training_data.append({"intent":"goodbye", "sentence":"good bye"})
training_data.append({"intent":"goodbye", "sentence":"talk to you soon"})
training_data.append({"intent":"goodbye", "sentence":"i have to go"})
training_data.append({"intent":"goodbye", "sentence":"i am going"})

Initializing pytorch model

ann = ANN(input_size, hidden_size, output_size)

Loss function

criterion = nn.NLLLoss()

Model is trained in pytorch with the iteration of these lines, last line backpropagate the loss.

output_p = ann(input)
loss = criterion(output_p,output)
loss.backward()

After some iterations model learned the task and stored what it learned in a file, 'ann.pt'.

It worked !

The model is tested with texts it didnt seen before. The 'test.py' use the learned model stored in 'ann.pt' to predict the intent.

ann=torch.load('ann.pt') #importing trained model

Here goes the prediction !

User:hello
greeting
User:good afternoon
greeting
User:see you
goodbye
User:

Try out with your own dataset

Delete the 'ann.pt' file and make changes in dataset like this

training_data.append({"intent":"your_intent","sentence:"corresponding sentence "})

Run the train and test program again

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Built with ❤️ by Gopi