Feature/interactions

python-lib/backend/fetch_api.py

@@ -14,7 +14,6 @@
 import time
 from dataiku.customwebapp import get_webapp_config
 from chart_formatters.lift_chart import LiftChartFormatter
-from glm_handler.dku_model_metrics import ModelMetricsCalculator
 from .api_utils import calculate_base_levels
 
 visual_ml_trainer = model_cache = model_deployer =relativities_calculator = None
@@ -381,12 +380,12 @@ def get_model_comparison_data():
 
 @fetch_api.route("/get_model_metrics", methods=["POST"])
 def get_model_metrics():
-    current_app.logger.info("Getting Model Metrics") 
     if is_local:
         return jsonify(dummy_model_metrics)
 
     loading_thread.join()
     request_json = request.get_json()
+    current_app.logger.info(f"Getting Model Metrics with request: {request_json}") 
 
 
     models = [request_json["model1"], request_json["model2"]]
@@ -397,9 +396,9 @@ def get_model_metrics():
 
     for i, model in enumerate(models, start=1):
         model_retriever = VisualMLModelRetriver(model)
-        mmc = ModelMetricsCalculator(model_retriever)
-        model_aic, model_bic, model_deviance = mmc.calculate_metrics()
-
+        model_aic = model_retriever.predictor._clf.aic_value
+        model_bic  = model_retriever.predictor._clf.bic_value
+        model_deviance = model_retriever.predictor._clf.deviance_value
         model_key = f"Model_{i}"
 
         metrics["models"][model_key] = {
@@ -448,12 +447,30 @@ def export_model():
             for i in range(max_len):
                 for variable in variables:
                     if i < len(variable_keys[variable]):
-                        value = variable_keys[variable][i]
+                        value = sorted(variable_keys[variable])[i]
                         csv_output += "{},{},,".format(value, relativities_dict[variable][value])
                     else:
                         csv_output += ",,,"
                 csv_output += "\n"
-
+
+            variable_stats = model_cache.get_model(model).get('variable_stats')
+            relativities_interaction = model_cache.get_model(model).get('relativities_interaction')
+
+            if len(relativities_interaction) > 0:
+                unique_interactions = relativities_interaction.groupby(['feature_1', 'feature_2']).count().reset_index()
+                for _, interaction in unique_interactions.iterrows():
+                    feature_1 = interaction['feature_1']
+                    feature_2 = interaction['feature_2']
+                    these_relativities = relativities_interaction[(relativities_interaction['feature_1']==feature_1) & (relativities_interaction['feature_2']==feature_2)]
+                    csv_output += "{} * {}\n\n".format(feature_1, feature_2)
+                    csv_output += ",,{}\n".format(feature_1)
+                    sorted_value_1 = sorted(list(set(these_relativities['value_1'])))
+                    csv_output += ",,{}\n{}".format(",".join([str(v) for v in sorted_value_1]), feature_2)
+                    sorted_value_2 = sorted(list(set(these_relativities['value_2'])))
+                    for value_2 in sorted_value_2:
+                        csv_output += ",{},{}\n".format(str(value_2), ",".join([str(these_relativities[(these_relativities['value_1']==value_1) & (these_relativities['value_2']==value_2)]['relativity'].iloc[0]/relativities_dict[feature_1][value_1]/relativities_dict[feature_2][value_2]) for value_1 in sorted_value_1]))
+                    csv_output += "\n"                    
+
             csv_data = csv_output.encode('utf-8')
 
         except KeyError as e:

python-lib/backend/local_config.py

@@ -115,7 +115,9 @@ def setup_dataiku_client():
         }
         }
 interaction_setup_params = {
-       "target_column": "ClaimAmount",
+    "target_column": "ClaimAmount",
+    "interactions": [{ "first": 'VehBrand', "second": 'VehPower' },
+                     { "first": 'Area', "second": 'DrivAge' }],
     "exposure_column":"Exposure",
     "distribution_function":"Tweedie",
     "link_function":"Logit",

python-lib/backend/model_cache.py

@@ -1,7 +1,6 @@
 from glm_handler.dku_relativites_calculator import RelativitiesCalculator
 from glm_handler.dku_model_deployer import ModelDeployer
 from glm_handler.glm_data_handler import GlmDataHandler
-from glm_handler.dku_model_metrics import ModelMetricsCalculator
 from time import time
 from logging_assist.logging import logger
 from glm_handler.glm_data_handler import GlmDataHandler
@@ -45,6 +44,7 @@ def setup_model_cache(global_dku_mltask, model_deployer):
                 base_values = relativities_calculator.get_base_values()
 
                 relativities = relativities_calculator.get_relativities_df()
+                relativities_interaction = relativities_calculator.get_relativities_interactions_df()
 
                 logger.info(f"relativites are: {relativities.to_dict()}")
                 relativities_dict = relativities_calculator.relativities
@@ -72,6 +72,7 @@ def setup_model_cache(global_dku_mltask, model_deployer):
                 # Store data in cache
                 model_cache.add_model(model_id, 
                                      relativities, 
+                                     relativities_interaction,
                                      model1_predicted_base,
                                      base_values,
                                      relativities_dict,
@@ -109,6 +110,8 @@ def update_model_cache(global_dku_mltask, model_cache):
             base_values = relativities_calculator.get_base_values()
 
             relativities = relativities_calculator.get_relativities_df()
+            relativities_interaction = relativities_calculator.get_relativities_interactions_df()
+
             relativities_dict = relativities_calculator.relativities
             variable_level_stats = VariableLevelStatsFormatter(
                 model_retriever, data_handler, relativities_calculator
@@ -132,6 +135,7 @@ def update_model_cache(global_dku_mltask, model_cache):
 
             model_cache.add_model(model_id, 
                                  relativities, 
+                                 relativities_interaction,
                                  model1_predicted_base,
                                  base_values,
                                  relativities_dict,