Fixed external training dataset parameter. Now updates orig_params in…

…stead of params

atomsci/ddm/pipeline/predict_from_model.py

@@ -156,7 +156,7 @@ def predict_from_model_file(model_path, input_df, id_col='compound_id', smiles_c
 
     pipe = mp.create_prediction_pipeline_from_file(pred_params, reload_dir=None, model_path=model_path)
     if external_training_data is not None:
-        pipe.params.dataset_key=external_training_data
+        pipe.orig_params.dataset_key=external_training_data
     pred_df = pipe.predict_full_dataset(input_df, contains_responses=has_responses, is_featurized=is_featurized,
                                         dset_params=pred_params, AD_method=AD_method, k=k, dist_metric=dist_metric,
                                         max_train_records_for_AD=max_train_records_for_AD, AD_return_NN=AD_return_NN)

atomsci/ddm/test/integrative/external_dataset/H1_graphconv.json

@@ -0,0 +1,19 @@
+{
+    "system": "LC",
+    "datastore": "False",
+    "save_results": "False",
+    "data_owner": "username",
+    "prediction_type": "regression",
+    "dataset_key": "../../test_datasets/H1_std.csv",
+    "id_col": "compound_id",
+    "smiles_col": "base_rdkit_smiles",
+    "response_cols": "pKi_mean",
+    "split_uuid": "002251a2-83f8-4511-acf5-e8bbc5f86677",
+    "previously_split": "True",
+    "uncertainty": "True",
+    "verbose": "True",
+    "transformers": "True",
+    "model_type": "NN",
+    "featurizer": "graphconv",
+    "result_dir": "./output"
+}

atomsci/ddm/test/integrative/external_dataset/test_external_dataset.py

@@ -0,0 +1,71 @@
+#!/usr/bin/env python
+
+import json
+import pandas as pd
+import os
+import sys
+
+import atomsci.ddm.pipeline.parameter_parser as parse
+from atomsci.ddm.pipeline import model_pipeline as mp
+from atomsci.ddm.pipeline import predict_from_model as pfm
+import pytest
+
+def clean():
+    """Clean test files"""
+    if "output" not in os.listdir():
+        os.mkdir("output")
+    for f in os.listdir("./output"):
+        if os.path.isfile("./output/"+f):
+            os.remove("./output/"+f)
+
+def test():
+    """Test AD index calculation: Curate data, fit model, and predict property for new compounds for each feature set"""
+
+    # Clean
+    # -----
+    clean()
+
+    # Run HyperOpt
+    # ------------
+    with open("H1_graphconv.json", "r") as f:
+        hp_params = json.load(f)
+
+    script_dir = parse.__file__.strip("parameter_parser.py").replace("/pipeline/", "")
+    python_path = sys.executable
+    hp_params["script_dir"] = script_dir
+    hp_params["python_path"] = python_path
+
+    params = parse.wrapper(hp_params)
+    if not os.path.isfile(params.dataset_key):
+        params.dataset_key = os.path.join(params.script_dir, params.dataset_key)
+
+    train_df = pd.read_csv(params.dataset_key)
+
+    pl = mp.ModelPipeline(params)
+    pl.train_model()
+
+    # this should raise an exception
+    with pytest.raises(Exception, match='Dataset file file_does_not_exist.csv does not exist'):
+        pred_df_file = pfm.predict_from_model_file(model_path=pl.params.model_tarball_path,
+                                            input_df=train_df[:10],
+                                            id_col="compound_id",
+                                            smiles_col="base_rdkit_smiles",
+                                            response_col="pKi_mean",
+                                            dont_standardize=True,
+                                            AD_method="z_score",
+                                            external_training_data='file_does_not_exist.csv')
+
+    # this should work
+    pred_df_file = pfm.predict_from_model_file(model_path=pl.params.model_tarball_path,
+                                        input_df=train_df[:10],
+                                        id_col="compound_id",
+                                        smiles_col="base_rdkit_smiles",
+                                        response_col="pKi_mean",
+                                        dont_standardize=True,
+                                        AD_method="z_score")
+    assert("AD_index" in pred_df_file.columns.values), 'Error: No AD_index column in pred_df_file'
+
+    clean()
+
+if __name__ == '__main__':
+    test()