Allow ignoring cache per-estimator, closes #16

fseval/pipeline/estimator.py

@@ -6,6 +6,7 @@
 from fseval.types import (
     AbstractEstimator,
     AbstractStorageProvider,
+    CacheUsage,
     IncompatibilityError,
     Task,
 )
@@ -17,7 +18,8 @@
 @dataclass
 class EstimatorConfig:
     estimator: Any = None  # must have _target_ of type BaseEstimator.
-    use_cache_if_available: bool = True
+    load_cache: CacheUsage = CacheUsage.allow
+    save_cache: CacheUsage = CacheUsage.allow
     # tags
     multioutput: Optional[bool] = None
     multioutput_only: Optional[bool] = None
@@ -36,7 +38,8 @@ class TaskedEstimatorConfig(EstimatorConfig):
     name: str = MISSING
     classifier: Optional[EstimatorConfig] = None
     regressor: Optional[EstimatorConfig] = None
-    use_cache_if_available: bool = True
+    load_cache: CacheUsage = CacheUsage.allow
+    save_cache: CacheUsage = CacheUsage.allow
     # tags
     multioutput: Optional[bool] = False
     multioutput_only: Optional[bool] = False
@@ -75,16 +78,31 @@ def _get_class_repr(cls, estimator):
         return f"{module_name}.{class_name}"
 
     def _load_cache(self, filename: str, storage_provider: AbstractStorageProvider):
+        if self.load_cache == CacheUsage.never:
+            self.logger.info("ignoring cache load completely.")
+            return
+
         restored = storage_provider.restore_pickle(filename)
         self.estimator = restored or self.estimator
         self._is_fitted = bool(restored)
 
+        if self.load_cache == CacheUsage.must:
+            assert self._is_fitted, (
+                "Cache usage was set to 'must' but loading cached estimator failed."
+                + " Pickle file might be corrupt or could not be found."
+            )
+
     def _save_cache(self, filename: str, storage_provider: AbstractStorageProvider):
-        storage_provider.save_pickle(filename, self.estimator)
+        if self.save_cache == CacheUsage.never:
+            self.logger.info("ignoring cache save completely.")
+            return
+        else:
+            storage_provider.save_pickle(filename, self.estimator)
+            # TODO check whether file was successfully saved.
 
     def fit(self, X, y):
         # don't refit if cache available and `use_cache_if_available` is enabled
-        if self._is_fitted and self.use_cache_if_available:
+        if self._is_fitted:
             self.logger.debug("using estimator from cache.")
             return self
 

fseval/storage_providers/wandb.py

@@ -29,8 +29,8 @@ class WandbStorageProvider(AbstractStorageProvider):
 
         run_id: Optional[str] - recover from a specific run id."""
 
-    resume: Optional[str] = None
     local_dir: Optional[str] = None
+    resume: Optional[str] = None
     entity: Optional[str] = None
     project: Optional[str] = None
     run_id: Optional[str] = None

fseval/types.py

@@ -11,6 +11,25 @@ class Task(Enum):
     classification = 2
 
 
+class CacheUsage(Enum):
+    """
+    Determines how cache usage is handled. In the case of **loading** caches:
+
+    - `allow`: program might use cache; if found and could be restored
+    - `must`: program should fail if no cache found
+    - `never`: program should not load cache even if found
+
+    When **saving** caches:
+    - `allow`: program might save cache; no fatal error thrown when fails
+    - `must`: program must save cache; throws error if fails (e.g. due to out of memory)
+    - `never`: program does not try to save a cached version
+    """
+
+    allow = 1
+    must = 2
+    never = 3
+
+
 class IncompatibilityError(Exception):
     ...
 

tests/integration/pipelines/test_rank_and_validate.py

@@ -144,6 +144,9 @@ def cfg(dataset, cv, resample, classifier, ranker, validator):
         n_bootstraps=2,
         n_jobs=None,
         all_features_to_select="range(1, min(50, p) + 1)",
+        upload_ranking_scores=True,
+        upload_validation_scores=True,
+        upload_best_scores=True,
     )
 
     cfg = OmegaConf.create(config.__dict__)