Merge pull request #54 from databricks-industry-solutions/small_doc_u…

…pdate

Small doc update

examples/foundation_daily.py

@@ -107,7 +107,7 @@ def transform_group(df):
 # MAGIC %md ### Models
 # MAGIC Let's configure a list of models we are going to apply to our time series for evaluation and forecasting. A comprehensive list of all supported models is available in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml). Look for the models where `model_type: foundation`; these are the foundation models we import from [chronos](https://github.com/amazon-science/chronos-forecasting/tree/main), [uni2ts](https://github.com/SalesforceAIResearch/uni2ts) and [moment](https://github.com/moment-timeseries-foundation-model/moment). Check their documentation for the detailed description of each model. 
 # MAGIC
-# MAGIC Foundation time series models are pretrained on millions or billions of time series. These models can produce analysis (i.e. forecasting, anomaly detection, classfication) on an unforeseen time series without training or tuning. You can modify the hyperparameters in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml) or overwrite the default values in [mmf_sa/forecasting_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/forecasting_conf.yaml).
+# MAGIC Foundation time series models are pretrained on millions or billions of time series. These models can produce analysis (i.e. forecasting, anomaly detection, classfication) on an unforeseen time series without training or tuning. You can modify the hyperparameters in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml) or overwrite the default values in [mmf_sa/forecasting_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/forecasting_conf.yaml). You can also introduce new hyperparameters that are supported by the base models. To do this, first add those hyperparameters under the model specification in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml). Then, include these hyperparameters inside the model instantiation which happens in the model pipeline script: e.g. `ChronosT5Tiny` class in [mmf_sa/models/chronosforecast/ChronosPipeline.py](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/chronosforecast/ChronosPipeline.py).
 
 # COMMAND ----------
 

examples/global_daily.py

@@ -108,7 +108,7 @@ def transform_group(df):
 # MAGIC %md ### Models
 # MAGIC Let's configure a list of models we are going to apply to our time series for evaluation and forecasting. A comprehensive list of all supported models is available in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml). Look for the models where `model_type: global`; these are the global models we import from [neuralforecast](https://github.com/Nixtla/neuralforecast). Check their documentation for the detailed description of each model. 
 # MAGIC
-# MAGIC Some of these models perform [hyperparameter optimization](https://nixtlaverse.nixtla.io/neuralforecast/examples/automatic_hyperparameter_tuning.html) on its own to search for the best parameters. You can specify the range of the search or fix the values in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml) or overwrite the default values in [mmf_sa/forecasting_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/forecasting_conf.yaml).
+# MAGIC Some of these models perform [hyperparameter optimization](https://nixtlaverse.nixtla.io/neuralforecast/examples/automatic_hyperparameter_tuning.html) on its own to search for the best parameters. You can specify the range of the search or fix the values in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml) or overwrite the default values in [mmf_sa/forecasting_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/forecasting_conf.yaml). You can also introduce new hyperparameters that are supported by the base models. To do this, first add those hyperparameters under the model specification in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml). Then, include these hyperparameters inside the model instantiation which happens in the model pipeline script: e.g. `NeuralFcAutoNBEATSx` class in [mmf_sa/models/neuralforecast/NeuralForecastPipeline.py](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/neuralforecast/NeuralForecastPipeline.py).
 
 # COMMAND ----------
 
@@ -162,7 +162,7 @@ def transform_group(df):
 # COMMAND ----------
 
 # MAGIC %md
-# MAGIC For global models, we train the model once using the training dataset excluding `backtest_months`. We then use the same fitted model to produce the as-if forecasts for all back testing periods. We take this approach to make sure that there is no data leakage. See how MMF implements backtesting in `backtest` method in [mmf_sa/models/abstract_model.py](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/abstract_model.py).
+# MAGIC For global models, we train the model once using the training dataset excluding `backtest_months`. We then use the same fitted model to produce the as-if forecasts for all back testing periods. We do this to make sure that there is no data leakage. See how MMF implements backtesting in `backtest` method in [mmf_sa/models/abstract_model.py](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/abstract_model.py). But the model that gets registered in Unity Catalog is trained using the full history.
 # MAGIC
 # MAGIC We store the as-if forecasts together with the actuals for each backtesting period, so you can construct any metric of your interest. We provide a few out-of-the-box metrics for you (e.g. smape), but the idea here is that you construct your own metrics reflecting your business requirements and evaluate models based on those. For example, maybe you care more about the accuracy of the near-horizon forecasts than the far-horizon ones. In such case, you can apply a decreasing wieght to compute weighted aggregated metrics.
 # MAGIC

examples/local_univariate_daily.py

@@ -113,7 +113,7 @@ def transform_group(df):
 # MAGIC %md ### Models
 # MAGIC Let's configure a list of models we are going to apply to our time series for evaluation and forecasting. A comprehensive list of all supported models is available in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml). Look for the models where `model_type: local`; these are the local models we import from [statsforecast](https://github.com/Nixtla/statsforecast), [r fable](https://cran.r-project.org/web/packages/fable/vignettes/fable.html) and [sktime](https://github.com/sktime/sktime). Check their documentations for the detailed description of each model. 
 # MAGIC
-# MAGIC Some of these models perform hyperparameter optimization ([statsforecast Automatic Forecasting](https://nixtlaverse.nixtla.io/statsforecast/index.html#automatic-forecasting)) on itself to search for the best parameters. For other models, you can modify the model hyperparameters in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml) or overwrite the default values in [mmf_sa/forecasting_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/forecasting_conf.yaml).
+# MAGIC Some of these models perform hyperparameter optimization ([statsforecast Automatic Forecasting](https://nixtlaverse.nixtla.io/statsforecast/index.html#automatic-forecasting)) on its own for some hyperparameters. For other hyperparameters or models, you can modify the hyperparameters in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml) or overwrite the default values in [mmf_sa/forecasting_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/forecasting_conf.yaml). You can also introduce new hyperparameters that are supported by the base models. To do this, first add those hyperparameters under the model specification in [mmf_sa/models/models_conf.yaml](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/models_conf.yaml). Then, include these hyperparameters inside the model instantiation which happens in the model pipeline script: e.g. `StatsFcAutoArima` class in [mmf_sa/models/statsforecast/StatsFcForecastingPipeline.py](https://github.com/databricks-industry-solutions/many-model-forecasting/blob/main/mmf_sa/models/statsforecast/StatsFcForecastingPipeline.py).
 
 # COMMAND ----------