Merge pull request #28 from databricks-industry-solutions/neuralforec…

…ast-update update neuralforecast
databricks-industry-solutions · May 17, 2024 · 1da747e · 1da747e
2 parents 54fee6d + 1bf90a2
commit 1da747e
Show file tree

Hide file tree

Showing 15 changed files with 1,166 additions and 649 deletions.
diff --git a/01_mmf_univariate_daily_demo.py → demo_global_daily.py b/01_mmf_univariate_daily_demo.py → demo_global_daily.py
@@ -77,28 +77,7 @@ def create_m4_df():
 # COMMAND ----------
 
 active_models = [
-    "StatsForecastBaselineWindowAverage",
-    #"StatsForecastBaselineSeasonalWindowAverage",
-    #"StatsForecastBaselineNaive",
-    #"StatsForecastBaselineSeasonalNaive",
-    "StatsForecastAutoArima",
-    #"StatsForecastAutoETS",
-    #"StatsForecastAutoCES",
-    #"StatsForecastAutoTheta",
-    #"StatsForecastTSB",
-    #"StatsForecastADIDA",
-    #"StatsForecastIMAPA",
-    #"StatsForecastCrostonClassic",
-    #"StatsForecastCrostonOptimized",
-    #"StatsForecastCrostonSBA",
-    "RFableArima",
-    #"RFableETS",
-    #"RFableNNETAR",
-    #"RFableEnsemble",
-    #"RDynamicHarmonicRegression",
-    "SKTimeTBats",
-    #"SKTimeLgbmDsDt",
-    #"NeuralForecastRNN",
+    "NeuralForecastRNN",
     #"NeuralForecastLSTM",
     #"NeuralForecastNBEATSx",
     #"NeuralForecastNHITS",
@@ -128,7 +107,8 @@ def create_m4_df():
     train_data="mmf_train",
     scoring_data="mmf_train",
     scoring_output=f"{catalog}.{db}.daily_scoring_output",
-    metrics_output=f"{catalog}.{db}.daily_metrics_output",
+    evaluation_output=f"{catalog}.{db}.daily_evaluation_output",
+    model_output=f"{catalog}.{db}",
     group_id="unique_id",
     date_col="ds",
     target="y",
@@ -147,16 +127,17 @@ def create_m4_df():
     active_models=active_models,
     experiment_path=f"/Shared/mmf_experiment",
     use_case_name="mmf",
+    accelerator="gpu",
 )
 
 # COMMAND ----------
 
-# MAGIC %md ### Metrics output
-# MAGIC In the metrics output table, the metrics for all backtest windows and all models are stored. This info can be used to monitor model performance or decide which models should be taken into the final aggregated forecast.
+# MAGIC %md ### Evaluation output
+# MAGIC In the evaluation output table, the evaluation for all backtest windows and all models are stored. This info can be used to monitor model performance or decide which models should be taken into the final aggregated forecast.
 
 # COMMAND ----------
 
-# MAGIC %sql select * from solacc_uc.mmf.daily_metrics_output order by unique_id, model, backtest_window_start_date
+# MAGIC %sql select * from solacc_uc.mmf.daily_evaluation_output order by unique_id, model, backtest_window_start_date
 
 # COMMAND ----------
 
@@ -178,7 +159,11 @@ def create_m4_df():
 
 # COMMAND ----------
 
-# MAGIC %sql delete from solacc_uc.mmf.daily_metrics_output
+# MAGIC %md ### Delete Tables
+
+# COMMAND ----------
+
+# MAGIC %sql delete from solacc_uc.mmf.daily_evaluation_output
 
 # COMMAND ----------
 

diff --git a/02_mm_rossmann_forecasting.py → demo_global_external_regressors_daily.py b/02_mm_rossmann_forecasting.py → demo_global_external_regressors_daily.py
@@ -59,28 +59,7 @@
 from forecasting_sa import run_forecast
 
 active_models = [
-    #"StatsForecastBaselineWindowAverage",
-    #"StatsForecastBaselineSeasonalWindowAverage",
-    #"StatsForecastBaselineNaive",
-    #"StatsForecastBaselineSeasonalNaive",
-    #"StatsForecastAutoArima",
-    #"StatsForecastAutoETS",
-    #"StatsForecastAutoCES",
-    #"StatsForecastAutoTheta",
-    #"StatsForecastTSB",
-    #"StatsForecastADIDA",
-    #"StatsForecastIMAPA",
-    #"StatsForecastCrostonClassic",
-    #"StatsForecastCrostonOptimized",
-    #"StatsForecastCrostonSBA",
-    #"RFableArima",
-    #"RFableETS",
-    #"RFableNNETAR",
-    #"RFableEnsemble",
-    #"RDynamicHarmonicRegression",
-    "SKTimeLgbmDsDt",
-    #"SKTimeTBats",
-    #"NeuralForecastRNN",
+    "NeuralForecastRNN",
     #"NeuralForecastLSTM",
     #"NeuralForecastNBEATSx",
     #"NeuralForecastNHITS",
@@ -95,7 +74,8 @@
     train_data=f"{catalog}.{db}.rossmann_train",
     scoring_data=f"{catalog}.{db}.rossmann_test",
     scoring_output=f"{catalog}.{db}.rossmann_scoring_output",
-    metrics_output=f"{catalog}.{db}.rossmann_metrics_output",
+    evaluation_output=f"{catalog}.{db}.rossmann_evaluation_output",
+    model_output=f"{catalog}.{db}",
     group_id="Store",
     date_col="Date",
     target="Sales",
@@ -120,7 +100,7 @@
 
 # COMMAND ----------
 
-# MAGIC %sql select * from solacc_uc.mmf.rossmann_metrics_output order by Store, model, backtest_window_start_date
+# MAGIC %sql select * from solacc_uc.mmf.rossmann_evaluation_output order by Store, model, backtest_window_start_date
 
 # COMMAND ----------
 
@@ -132,7 +112,11 @@
 
 # COMMAND ----------
 
-# MAGIC %sql delete from solacc_uc.mmf.rossmann_metrics_output
+# MAGIC %md ### Delete Tables
+
+# COMMAND ----------
+
+# MAGIC %sql delete from solacc_uc.mmf.rossmann_evaluation_output
 
 # COMMAND ----------
 

diff --git a/01_mmf_univariate_monthly_demo.py → demo_global_monthly.py b/01_mmf_univariate_monthly_demo.py → demo_global_monthly.py
@@ -88,28 +88,7 @@ def transform_group(df):
 # COMMAND ----------
 
 active_models = [
-    "StatsForecastBaselineWindowAverage",
-    #"StatsForecastBaselineSeasonalWindowAverage",
-    #"StatsForecastBaselineNaive",
-    #"StatsForecastBaselineSeasonalNaive",
-    "StatsForecastAutoArima",
-    #"StatsForecastAutoETS",
-    #"StatsForecastAutoCES",
-    #"StatsForecastAutoTheta",
-    #"StatsForecastTSB",
-    #"StatsForecastADIDA",
-    #"StatsForecastIMAPA",
-    #"StatsForecastCrostonClassic",
-    #"StatsForecastCrostonOptimized",
-    #"StatsForecastCrostonSBA",
-    "RFableArima",
-    #"RFableETS",
-    #"RFableNNETAR",
-    #"RFableEnsemble",
-    #"RDynamicHarmonicRegression",
-    "SKTimeTBats",
-    #"SKTimeLgbmDsDt",
-    #"NeuralForecastRNN",
+    "NeuralForecastRNN",
     #"NeuralForecastLSTM",
     #"NeuralForecastNBEATSx",
     #"NeuralForecastNHITS",
@@ -139,7 +118,8 @@ def transform_group(df):
     train_data="mmf_train",
     scoring_data="mmf_train",
     scoring_output=f"{catalog}.{db}.monthly_scoring_output",
-    metrics_output=f"{catalog}.{db}.monthly_metrics_output",
+    evaluation_output=f"{catalog}.{db}.monthly_evaluation_output",
+    model_output=f"{catalog}.{db}",
     group_id="unique_id",
     date_col="date",
     target="y",
@@ -162,16 +142,16 @@ def transform_group(df):
 
 # COMMAND ----------
 
-# MAGIC %md ### Metrics output
-# MAGIC In the metrics output table, the metrics for all backtest windows and all models are stored. This info can be used to monitor model performance or decide which models should be taken into the final aggregated forecast.
+# MAGIC %md ### Evaluation Output
+# MAGIC In the evaluation output table, the evaluation for all backtest windows and all models are stored. This info can be used to monitor model performance or decide which models should be taken into the final aggregated forecast.
 
 # COMMAND ----------
 
-# MAGIC %sql select * from solacc_uc.mmf.monthly_metrics_output order by unique_id, model, backtest_window_start_date
+# MAGIC %sql select * from solacc_uc.mmf.monthly_evaluation_output order by unique_id, model, backtest_window_start_date
 
 # COMMAND ----------
 
-# MAGIC %md ### Forecast output
+# MAGIC %md ### Forecast Output
 # MAGIC In the Forecast output table, the final forecast for each model and each time series is stored. 
 
 # COMMAND ----------
@@ -180,7 +160,7 @@ def transform_group(df):
 
 # COMMAND ----------
 
-# MAGIC %md ### Final Ensemble Output
+# MAGIC %md ### Ensemble Output
 # MAGIC In the final ensemble output table, we store the averaged forecast. The models which meet the threshold defined using the ensembling parameters are taken into consideration
 
 # COMMAND ----------
@@ -189,7 +169,11 @@ def transform_group(df):
 
 # COMMAND ----------
 
-# MAGIC %sql delete from solacc_uc.mmf.monthly_metrics_output
+# MAGIC %md ### Delete Tables
+
+# COMMAND ----------
+
+# MAGIC %sql delete from solacc_uc.mmf.monthly_evaluation_output
 
 # COMMAND ----------