known in advance and accuracy tracking

YoutubeForecastMaker/conf/base/parameters.yml

@@ -41,12 +41,34 @@ deploy_forecast:
       multiseries_id_columns: [title]
       use_time_series: True
       # Forecast/Feature Derivation Window start/ends must be a multiple of the time step
-      feature_derivation_window_start: -21 
+      feature_derivation_window_start: -72
       feature_derivation_window_end: 0
       forecast_window_start: 3 
       forecast_window_end: 36
       # TODO: what other configurations should we add? Should we add and leave empty?
       #   - Known in advance?
+      default_to_known_in_advance: true
+    feature_settings_config:
+        - feature_name: video_id
+          known_in_advance: true
+        - feature_name: publishedAt
+          known_in_advance: true
+        - feature_name: channelId
+          known_in_advance: true
+        - feature_name: description
+          known_in_advance: true
+        - feature_name: categoryId
+          known_in_advance: true
+        - feature_name: channelTitle
+          known_in_advance: true
+        - feature_name: tags
+          known_in_advance: true
+        - feature_name: duration
+          known_in_advance: true
+        - feature_name: madeForKids
+          known_in_advance: true
+        - feature_name: association_id
+          known_in_advance: true
     advanced_options_config:
       seed: 42
   registered_model:
@@ -66,19 +88,54 @@ deploy_forecast:
         runLeakageRemovedFeatureList: True
       trigger:
         type: accuracy_decline
+        schedule: 
+          minute: 
+            - 0
+          hour:
+            - 0
+          dayOfMonth:
+            - "*"
+          month:
+            - "*"
+          dayOfWeek:
+            - "*"
         statusDeclinesToWarning: true,
         statusDeclinesToFailing: false
         statusStillInDecline: false
+      projectOptions:
+        cvMethod: RandomCV
+        validationType: CV
+        reps: null
+        validationPct: null
+        holdoutPct: null
+        metric: RMSE
       featureListStrategy: informative_features
       projectOptionsStrategy: same_as_champion
       modelSelectionStrategy: autopilot_recommended
     action: model_replacement
   batch_prediction_job_definition:
     batch_prediction_job:
-      intake_settings: 
+      num_concurrent: 3
+      intake_settings:
         type: dataset
-        dataset: 
-    enabled: True
+    enabled: true
+    schedule: 
+          minute: 
+            - 0
+          hour:
+            - 0
+            - 3
+            - 6
+            - 9
+            - 12
+            - 15
+            - 18
+          dayOfMonth:
+            - "*"
+          month:
+            - "*"
+          dayOfWeek:
+            - "*"
     name: Batch Prediction Job for Retraining
 
 

YoutubeForecastMaker/src/YoutubeForecastMaker/pipelines/deploy_forecast/nodes.py

@@ -13,7 +13,7 @@
 from datarobot.models.use_cases.utils import UseCaseLike
 from datarobot import Dataset 
 
-from datarobotx.idp.batch_predictions import update_or_create_batch_prediction_job
+from datarobotx.idp.batch_predictions import get_update_or_create_batch_prediction_job
 
 if TYPE_CHECKING:
     import tempfile
@@ -176,9 +176,10 @@ def setup_batch_prediction_job_definition(
     """
     dr.Client(token=token, endpoint=endpoint)  # type: ignore
 
-    batch_prediction_job["intake_settings"]["dataset"] = dr.Dataset.get(dataset_id=dataset_id)
+    batch_prediction_job["intake_settings"]["datasetId"] = dataset_id
+    batch_prediction_job["deploymentId"] = deployment_id
 
-    update_or_create_batch_prediction_job(endpoint=endpoint,
+    get_update_or_create_batch_prediction_job(endpoint=endpoint,
                                           token=token,
                                           deployment_id=deployment_id,
                                           batch_prediction_job=batch_prediction_job,

YoutubeForecastMaker/src/YoutubeForecastMaker/pipelines/deploy_forecast/pipeline.py

@@ -19,7 +19,7 @@
     get_or_create_registered_leaderboard_model_version,
 )
 from datarobotx.idp.use_cases import get_or_create_use_case
-from datarobotx.idp.retraining_policies import update_or_create_retraining_policy
+from datarobotx.idp.retraining_policies import get_update_or_create_retraining_policy
 
 from .nodes import (ensure_deployment_settings, 
                     put_forecast_distance_into_registered_model_name,
@@ -58,6 +58,7 @@ def create_pipeline(**kwargs) -> Pipeline:
                 "dataset_id": "preprocessed_timeseries_data_id",
                 "analyze_and_model_config": "params:project.analyze_and_model_config",
                 "datetime_partitioning_config": "params:project.datetime_partitioning_config",
+                # "feature_settings_config": "params:project.feature_settings_config",
                 "advanced_options_config": "params:project.advanced_options_config",
                 "use_case": "use_case_id",
             },
@@ -139,20 +140,22 @@ def create_pipeline(**kwargs) -> Pipeline:
                 "dataset_id": "preprocessed_timeseries_data_id",
                 "enabled": "params:batch_prediction_job_definition.enabled",
                 "name": "params:batch_prediction_job_definition.name",
-                "batch_prediction_job": "params:batch_prediction_job_definition.batch_prediction_job"
+                "batch_prediction_job": "params:batch_prediction_job_definition.batch_prediction_job",
+                "schedule": "params:batch_prediction_job_definition.schedule"
             },
             outputs=None
         ),
-        node(
+       node(
             name="set_up_retraining_job",
-            func=update_or_create_retraining_policy,
+            func=lambda endpoint, token, deployment_id, name, dataset_id, retraining_settings: get_update_or_create_retraining_policy(
+                    endpoint, token, deployment_id, name, dataset_id, **retraining_settings),
             inputs={
                 "endpoint": "params:credentials.datarobot.endpoint",
                 "token": "params:credentials.datarobot.api_token",
                 "deployment_id": "deployment_id",
                 "name": "params:retraining_policy.name",
                 "dataset_id": "preprocessed_timeseries_data_id",
-                "kwargs": "params:retraining_policy.kwargs"
+                "retraining_settings": "params:retraining_policy.retraining_settings"
             },
             outputs=None
         )

YoutubeForecastMaker/src/YoutubeForecastMaker/pipelines/deploy_streamlit_app/nodes.py

@@ -136,7 +136,7 @@ def log_outputs(
 
     base_url = urljoin(endpoint, "/")
     project_url = base_url + "projects/{project_id}/models/{model_id}/"
-    deployment_url = base_url + "console/{deployment_id}/overview"
+    deployment_url = base_url + "deployments/{deployment_id}/overview"
     application_url = base_url + "custom_applications/{application_id}/"
 
     logger = logging.getLogger(__name__)