Merge pull request #83 from isi-usc-edu/johnp/miniML-fixes

fixes issue #74 and #75

BubbleML/miniML/README.md

@@ -33,11 +33,9 @@ options:
 Example call:
 
 ```bash
-./miniML.py --ham_features_file TESTING_ONLY_Hamiltonian_features.csv --config_file miniML_config.json --solver_uuid bd63d0e0-a681-11ef-b4bd-6f1bb9e0689f --solver_labels_file TESTING_ONLY_solver.ccsdt.labels.csv --verbose
+./miniML.py -v --ham_features_file ../../Hamiltonian_features/experimental/fast_double_factorization_features/Hamiltonian_features.csv --config_file miniML_config.json --solver_uuid 16537433-9f4c-4eae-a65d-787dc3b35b59 --solver_labels_file ../../scripts/solver_labels.DMRG_Niagara_cluster_lowest_energy.16537433-9f4c-4eae-a65d-787dc3b35b59.csv --verbose
 ```
 
-Note that the `solver_uuid` is contrived for this example.  The solver is CCSDT in this example.
+Note that the `solver_uuid` is specific to the DMRG algorithm running on a specific compute platform in this example.
 
 If the `--verbose` flag is included, then a `plot_<solver_uuid, datestamp>.png` plot and the `probs<solver_uuid, datestamp>.csv` file will be generated as artifacts.
-
-TODO: As we are integrating, use the `TESTING_ONLY_Hamiltonian_features.csv` and `TESTING_ONLY_solver.ccsdt.labels.csv` files.  As more data is collated, we should point to `../../Hamiltonian_features/experimental/fast_double_factorization/Hamiltonian_features.csv` and an appropriately aligned `solver.<solver_uuid>.labels.csv` files.

BubbleML/miniML/miniML.py

@@ -140,15 +140,30 @@ def trainML(
         )
         logging.info('Percent of solvable space: ', str(ml_solvability_ratio))
 
+        timestamp = datetime.datetime.now().strftime("%Y-%m-%d-%H-%M")
+
+
         #explain all the predictions in the test set
         plt.figure()
         explainer = shap.KernelExplainer(model.predict_proba, X_train)
         shap_values = explainer.shap_values(X_train)
         class_index = 1
         shap.initjs()
-        shap.summary_plot(shap_values[1],features=X.columns,plot_type="bar")
+
+        shap.summary_plot(
+            shap_values[1],
+            features=X.columns,
+            plot_type="bar"
+        )
+
         #shap.force_plot(explainer.expected_value[class_index], shap_values[class_index], X_train, matplotlib=True, show=False)
 
+        plt.savefig(
+            f"shap_summary_plot_solver={solver_uuid}_{timestamp}.png",
+            format="png"
+        )
+
+
         if verbose:
             # print to file
             y_pred = model.predict(X_train)
@@ -164,7 +179,6 @@ def trainML(
                     "prob_class_1": probs[:,1]
                 }
             )
-            timestamp = datetime.datetime.now().strftime("%Y-%m-%d-%H-%M")
             probs_file_name = f"probs_solver={solver_uuid}_{timestamp}.csv"
             df.to_csv(probs_file_name, index=False)
             logging.info(f"wrote probs to file {probs_file_name}.")