def H1_init():
"""
Test full model pipeline: Curate data, fit model, and predict property for new compounds
"""
# Clean
# -----
integrative_utilities.clean_fit_predict()
clean('H1')
# Curate
# ------
H1_curate()
def init():
"""
Test full model pipeline: Curate data, fit model, and predict property for new compounds
"""
# Clean
# -----
integrative_utilities.clean_fit_predict()
clean()
# Download
# --------
download()
# Curate
# ------
curate()
def test():
"""
Test full model pipeline: Curate data, fit model, and predict property for new compounds
"""
# Clean
# -----
integrative_utilities.clean_fit_predict()
clean()
# Download
# --------
download()
# Curate
# ------
curate()
# Train model
# -----------
# Read parameter JSON file
with open('config_delaney_train_NN.json') as f:
config = json.loads(f.read())
# Parse parameters
params = parse.wrapper(config)
# Create model pipeline
model = mp.ModelPipeline(params)
# Train model
model.train_model()
# Get uuid and reload directory
# -----------------------------
uuid = integrative_utilities.get_subdirectory(
'result/delaney-processed_curated_fit/NN_graphconv_scaffold_regression'
)
reload_dir = 'result/delaney-processed_curated_fit/NN_graphconv_scaffold_regression/' + uuid
# Check training statistics
# -------------------------
integrative_utilities.training_statistics_file(reload_dir, 'test', 0.6)
# Make prediction parameters
# --------------------------
# Read prediction parameter JSON file
with open('config_delaney_predict_NN.json', 'r') as f:
predict_parameters_dict = json.loads(f.read())
# Set transformer key here because model uuid is not known before fit
predict_parameters_dict[
'transformer_key'] = reload_dir + 'transformers.pkl'
predict_parameters = parse.wrapper(predict_parameters_dict)
# Load second test set
# --------------------
data = pd.read_csv('delaney-processed_curated_external.csv')
# Select columns and rename response column
data = data[[
predict_parameters.id_col, predict_parameters.smiles_col,
predict_parameters.response_cols[0]
]]
data = data.rename(
columns={predict_parameters.response_cols[0]: 'experimental_values'})
# Make prediction pipeline
# ------------------------
pp = mp.create_prediction_pipeline_from_file(predict_parameters,
reload_dir)
# Predict
# -------
predict = pp.predict_on_dataframe(data)
# Check predictions
# -----------------
assert (predict['pred'].shape[0] == 117
), 'Error: Incorrect number of predictions'
assert (np.all(np.isfinite(
predict['pred'].values))), 'Error: Predictions are not numbers'
# Save predictions with experimental values
# -----------------------------------------
predict.reset_index(level=0, inplace=True)
combined = pd.merge(data,
predict,
on=predict_parameters.id_col,
how='inner')
combined.to_csv('delaney-processed_curated_predict.csv')
assert (os.path.isfile('delaney-processed_curated_predict.csv')
and os.path.getsize('delaney-processed_curated_predict.csv') > 0
), 'Error: Prediction file not created'