remove settings.OPENPREDICT_DATA_DIR to use get_openpredict_dir() fct

.github/workflows/publish-package.yml

@@ -2,8 +2,8 @@ name: Publish package
 # Publish to PyPI when new release on GitHub, if tests pass
 on:
   workflow_dispatch:
-  release:
-    types: [created]
+  # release:
+  #   types: [created]
 
 jobs:
 

src/drkg_model/download.py

@@ -3,7 +3,7 @@
 
 import os
 
-from trapi_predict_kit.utils import get_openpredict_dir
+from openpredict_model.utils import get_openpredict_dir
 
 
 # Downloading 500M kgpredict external dependency to avoid to have to commit it to dvc

src/drkg_model/predict.py

@@ -7,9 +7,10 @@
 import pandas as pd
 import torch as th
 import torch.nn.functional as fn
+from trapi_predict_kit.utils import get_entities_labels
 
 from drkg_model.download import download
-from trapi_predict_kit.utils import get_entities_labels, get_openpredict_dir
+from openpredict_model.utils import get_openpredict_dir
 
 # Predict drug repurposing based on the DRKG (drug repurposing KG) by Arif Yilmaz
 # TODO: cannot be integrated to TRAPI because it returns negative scores

src/openpredict_model/api.py

@@ -6,11 +6,11 @@
 
 from fastapi import APIRouter, File, Query, UploadFile
 from rdflib import Graph
-from trapi_predict_kit.rdf_utils import retrieve_features, retrieve_models
 
 from openpredict_model.evidence_path.predict import do_evidence_path
 from openpredict_model.explain_shap.explain_shap import get_explanations
 from openpredict_model.train import add_embedding
+from openpredict_model.utils import retrieve_features, retrieve_models
 
 
 class FeatureTypesDrugs(str, Enum):

src/openpredict_model/evidence_path/predict.py

@@ -4,17 +4,17 @@
 import pandas as pd
 from gensim.models import KeyedVectors
 
-from trapi_predict_kit.config import settings
 from openpredict_model.evidence_path.train import getQuantiles, path_weight_product
+from openpredict_model.utils import get_openpredict_dir
 
 ## Evidence path for OpenPredict model by Elif
 
-df_op = pd.read_csv(f"{settings.OPENPREDICT_DATA_DIR}/resources/openpredict-omim-drug.csv")
+df_op = pd.read_csv(f"{get_openpredict_dir()}/resources/openpredict-omim-drug.csv")
 
 drug_fp_vectors = KeyedVectors.load_word2vec_format(
-    f'{settings.OPENPREDICT_DATA_DIR}/embedding/drugs_fp_embed.txt', binary=False)
+    f'{get_openpredict_dir()}/embedding/drugs_fp_embed.txt', binary=False)
 disease_hp_vectors = KeyedVectors.load_word2vec_format(
-    f'{settings.OPENPREDICT_DATA_DIR}/embedding/disease_hp_embed.txt', binary=False)
+    f'{get_openpredict_dir()}/embedding/disease_hp_embed.txt', binary=False)
 
 df_op = df_op.rename(columns={'omimid': 'disease_id', 'drugid': 'drug_id'})
 df_op.disease_id = df_op.disease_id.astype(str)
@@ -32,7 +32,7 @@ def generate_paths_for_apair(drug, disease, drug_emb_vectors, disease_emb_vector
     (threshold_drug,threshold_disease) =getQuantiles(drug_emb_vectors, disease_emb_vectors, threshold_drugs)
 
     if(features_drug is not None) :
-         filtered_embedding_drugs = KeyedVectors.load_word2vec_format(f'{settings.OPENPREDICT_DATA_DIR}/evidence-path-model/feature_{str(features_drug)}.txt', binary=False)
+         filtered_embedding_drugs = KeyedVectors.load_word2vec_format(f'{get_openpredict_dir()}/evidence-path-model/feature_{str(features_drug)}.txt', binary=False)
          similarDrugs = filtered_embedding_drugs.most_similar(drug, topn=100)
          (threshold_drug,threshold_disease) =getQuantiles(filtered_embedding_drugs, disease_emb_vectors, threshold_drugs)
     else :
@@ -54,9 +54,9 @@ def generate_paths_for_apair(drug, disease, drug_emb_vectors, disease_emb_vector
     (threshold_drug,threshold_disease) =getQuantiles(drug_emb_vectors, disease_emb_vectors, threshold_diseases)
 
 
-    # TODO: USE settings.OPENPREDICT_DATA_DIR instead of lucky relative path
+    # TODO: USE get_openpredict_dir() instead of lucky relative path
     if(features_disease is not None) :
-        filtered_embedding_diseases = KeyedVectors.load_word2vec_format(f'{settings.OPENPREDICT_DATA_DIR}/evidence-path-model/feature_{str(features_disease)}.txt', binary=False)
+        filtered_embedding_diseases = KeyedVectors.load_word2vec_format(f'{get_openpredict_dir()}/evidence-path-model/feature_{str(features_disease)}.txt', binary=False)
         # filtered_embedding_diseases = KeyedVectors.load_word2vec_format(f'openpredict/data/embedding/feature_specific_embeddings_KG/feature_{str(features_disease)}.txt', binary=False)
         similarDiseases = filtered_embedding_diseases.most_similar(disease, topn=100)
         (threshold_drug,threshold_disease) =getQuantiles(drug_fp_vectors, filtered_embedding_diseases, threshold_diseases)

src/openpredict_model/evidence_path/train.py

@@ -3,16 +3,15 @@
 import pandas as pd
 from gensim.models import KeyedVectors
 
-from trapi_predict_kit.config import settings
-from openpredict_model.utils import load_features_embeddings
+from openpredict_model.utils import load_features_embeddings, get_openpredict_dir
 
 ## Evidence path for OpenPredict model by Elif
 
 
 drug_fp_vectors = KeyedVectors.load_word2vec_format(
-    f'{settings.OPENPREDICT_DATA_DIR}/embedding/drugs_fp_embed.txt', binary=False)
+    f'{get_openpredict_dir()}/embedding/drugs_fp_embed.txt', binary=False)
 disease_hp_vectors = KeyedVectors.load_word2vec_format(
-    f'{settings.OPENPREDICT_DATA_DIR}/embedding/disease_hp_embed.txt', binary=False)
+    f'{get_openpredict_dir()}/embedding/disease_hp_embed.txt', binary=False)
 
 
 ###############################################

src/openpredict_model/explain_shap/explain_shap.py

@@ -5,9 +5,10 @@
 import pandas as pd
 import shap as shap
 import sklearn
+from trapi_predict_kit.utils import get_entities_labels
 
-from trapi_predict_kit.utils import get_entities_labels, get_openpredict_dir
 from openpredict_model.predict import query_omim_drugbank_classifier
+from openpredict_model.utils import get_openpredict_dir
 
 # XPREDICT framework may be used to add explanation features to drug repositioning applications
 

src/openpredict_model/generate_disease_features.py

@@ -5,7 +5,7 @@
 import numpy as np
 import pandas as pd
 
-from trapi_predict_kit.config import settings
+from openpredict_model.utils import get_openpredict_dir
 
 # Standalone script to generate diseases features
 
@@ -16,7 +16,7 @@ def fasta2seq(lines):
 
 def download():
     """Download a jar file required for the processing in data/lib/"""
-    if not os.path.exists(settings.OPENPREDICT_DATA_DIR / "lib" / "sml-toolkit-0.9.jar" ):
+    if not os.path.exists(get_openpredict_dir() / "lib" / "sml-toolkit-0.9.jar" ):
         print("sml-toolkit-0.9.jar not present, downloading it")
         try:
             os.system(f'mkdir -p data/lib')

src/openpredict_model/predict.py

@@ -5,12 +5,11 @@
 import pandas as pd
 
 from trapi_predict_kit import load
-from trapi_predict_kit.config import settings
 from trapi_predict_kit.decorators import trapi_predict
 from trapi_predict_kit.predict_output import PredictOptions, PredictOutput
 from trapi_predict_kit.utils import get_entities_labels, get_entity_types, log
 from openpredict_model.train import createFeaturesSparkOrDF
-from openpredict_model.utils import load_features_embeddings, similarity_embeddings
+from openpredict_model.utils import load_features_embeddings, similarity_embeddings, get_openpredict_dir
 
 trapi_nodes = {
     "biolink:Disease": {
@@ -151,7 +150,7 @@ def query_omim_drugbank_classifier(input_curie, model_id):
 
     # TODO: should we update this file too when we create new runs?
     drugDiseaseKnown = pd.read_csv(
-        os.path.join(settings.OPENPREDICT_DATA_DIR, 'resources', 'openpredict-omim-drug.csv'), delimiter=',')
+        os.path.join(get_openpredict_dir(), 'resources', 'openpredict-omim-drug.csv'), delimiter=',')
     drugDiseaseKnown.rename(
         columns={'drugid': 'Drug', 'omimid': 'Disease'}, inplace=True)
     drugDiseaseKnown.Disease = drugDiseaseKnown.Disease.astype(str)

src/openpredict_model/train.py

@@ -15,17 +15,18 @@
 
 from trapi_predict_kit import save
 from trapi_predict_kit.config import settings
-from trapi_predict_kit.rdf_utils import get_run_id
 from trapi_predict_kit.utils import log
 
+from openpredict_model.utils import get_run_id, get_openpredict_dir
+
 cli = typer.Typer(help="Training for OpenPredict model")
 
 
 
 # @is_fairstep(label='Prepare known drug-disease associations dictionary')
 def get_known_associations():
     drugDiseaseKnown = pd.read_csv(
-        os.path.join(settings.OPENPREDICT_DATA_DIR, 'resources', 'openpredict-omim-drug.csv'),
+        os.path.join(get_openpredict_dir(), 'resources', 'openpredict-omim-drug.csv'),
         delimiter=','
     )
     drugDiseaseKnown.rename(
@@ -46,20 +47,20 @@ def get_drug_disease_features(from_model_id: str):
         drugfeatfiles = ['drugs-fingerprint-sim.csv', 'drugs-se-sim.csv',
                          'drugs-ppi-sim.csv', 'drugs-target-go-sim.csv', 'drugs-target-seq-sim.csv']
         diseasefeatfiles = ['diseases-hpo-sim.csv',  'diseases-pheno-sim.csv']
-        drugfeatfiles = [os.path.join(settings.OPENPREDICT_DATA_DIR,
+        drugfeatfiles = [os.path.join(get_openpredict_dir(),
             baseline_features_folder, fn) for fn in drugfeatfiles]
-        diseasefeatfiles = [os.path.join(settings.OPENPREDICT_DATA_DIR,
+        diseasefeatfiles = [os.path.join(get_openpredict_dir(),
             baseline_features_folder, fn) for fn in diseasefeatfiles]
         # baseline_features_folder = "data/baseline_features/"
         # TODO: Translator IDs version (MONDO & CHEBI)
         drug_df, disease_df = mergeFeatureMatrix(
             drugfeatfiles, diseasefeatfiles)
     else:
         print(type(from_model_id))
-        print(f"📥 Loading the features tensor from {settings.OPENPREDICT_DATA_DIR}/features/{str(from_model_id)}_features.pickle")
+        print(f"📥 Loading the features tensor from {get_openpredict_dir()}/features/{str(from_model_id)}_features.pickle")
 
         (drug_df, disease_df) = pickle.load(open(
-            f"{settings.OPENPREDICT_DATA_DIR}/features/{str(from_model_id)}_features.pickle",
+            f"{get_openpredict_dir()}/features/{str(from_model_id)}_features.pickle",
             "rb"
         ))
     print("Drug Features ", drug_df.columns.levels[0])
@@ -656,7 +657,7 @@ def train_model(from_model_id: str = 'openpredict_baseline'):
     #     pickle.dump(
     #         (drug_df, disease_df),
     #         open(
-    #             f"{settings.OPENPREDICT_DATA_DIR}/features/{from_model_id}_features.pickle",
+    #             f"{get_openpredict_dir()}/features/{from_model_id}_features.pickle",
     #             "wb"
     #         )
     #     )
@@ -708,9 +709,9 @@ def add_embedding(
     # TODO: now also save the feature dataframe for each run to be able to add embedding to any run?
     # Or can we just use the models/run_id.pickle file instead of having 2 files for 1 run?
     print('📥 Loading features file: ' +
-          f"{settings.OPENPREDICT_DATA_DIR}/features/{from_model_id}_features.pickle")
+          f"{get_openpredict_dir()}/features/{from_model_id}_features.pickle")
     (drug_df, disease_df) = pickle.load(open(
-        f"{settings.OPENPREDICT_DATA_DIR}/features/{from_model_id}_features.pickle",
+        f"{get_openpredict_dir()}/features/{from_model_id}_features.pickle",
         "rb"
     ))
 
@@ -761,7 +762,7 @@ def add_embedding(
     pickle.dump(
         (drug_df, disease_df),
         open(
-            f"{settings.OPENPREDICT_DATA_DIR}/features/{run_id}_features.pickle",
+            f"{get_openpredict_dir()}/features/{run_id}_features.pickle",
             "wb"
         )
     )