starting implementing deepar with mlflow

climate_health/external/external_model.py

@@ -196,6 +196,9 @@ def predict(self, future_data: IsSpatioTemporalDataSet[FeatureType]) -> IsSpatio
             df = future_data.to_pandas()
             df['disease_cases'] = np.nan
 
+            # todo: instead of using saved state for historic data, get histori data in as argument to predict
+            # send historic data and future data as two seperate data sets to model
+
             new_pd = self._adapt_data(df)
             if self.is_lagged:
                 new_pd = pd.concat([self._saved_state, new_pd]).sort_values(['location', 'time_period'])
@@ -413,7 +416,7 @@ def get_model_from_mlproject_file(mlproject_file):
         adapters = config.get('adapters', None)
         allowed_data_types = {'HealthData': HealthData}
         data_type = allowed_data_types.get(config.get('data_type', None), None)
-        return ExternalMLflowModel(mlproject_file, name=name, adapters=adapters, data_type=data_type,
+        return ExternalMLflowModel(mlproject_file.parent, name=name, adapters=adapters, data_type=data_type,
                                    working_dir=Path(mlproject_file).parent)
 
 

climate_health/external/mlflow.py

@@ -32,7 +32,7 @@ def __init__(self, model_path: str, name: str=None, adapters=None, working_dir="
     @property
     def name(self):
         return self._name
-    
+
     def __call__(self):
         return self
 
@@ -55,7 +55,7 @@ def train(self, train_data: DataSet, extra_args=None):
         response = mlflow.projects.run(str(self.model_path), entry_point="train",
                                        parameters={
                                            "train_data": str(train_file_name),
-                                           "model_output_file": str(self._model_file_name)
+                                           "model": str(self._model_file_name)
                                        },
                                        build_image=True)
         self._saved_state = new_pd

external_models/deepar/MLproject

@@ -0,0 +1,22 @@
+name: My Project
+
+# python_env: python_env.yaml
+# or
+# conda_env: my_env.yaml
+# or
+# docker_env:
+#    image:  mlflow-docker-example
+
+entry_points:
+  train:
+    parameters:
+      train_data: path
+      model: path
+    command: "ch_modelling train {train_data} {model}"
+  predict:
+    parameters:
+      model: path
+      historic_data: path
+      future_data: path
+      out_file: path
+    command: "ch_modelling predict {model} {historic_data} {future_data} {out_file}"

requirements_dev.txt

@@ -17,3 +17,4 @@ earthengine-api
 python-dotenv
 myst_parser
 furo
+virtualenv