refactor(code): improve abstractions (#28)

Author: fmind

src/bikes/io/__init__.py

@@ -1 +1 @@
-"""Components related to external operations."""
+"""Components related to external operations (inputs and outputs)."""

src/bikes/io/datasets.py

@@ -69,11 +69,12 @@ class ParquetReader(Reader):
     KIND: T.Literal["ParquetReader"] = "ParquetReader"
 
     path: str
+    backend: T.Literal["pyarrow", "numpy_nullable"] = "pyarrow"
 
     @T.override
     def read(self) -> pd.DataFrame:
         # can't limit rows at read time
-        data = pd.read_parquet(self.path)
+        data = pd.read_parquet(self.path, dtype_backend="pyarrow")
         if self.limit is not None:
             data = data.head(self.limit)
         return data

src/bikes/io/services.py

@@ -199,7 +199,7 @@ def run_context(self, run_config: RunConfig) -> T.Generator[mlflow.ActiveRun, No
             run (str): run parameters.
 
         Yields:
-            T.Generator[mlflow.ActiveRun, None, None]: active run context. Will be closed as the end of context.
+            T.Generator[mlflow.ActiveRun, None, None]: active run context. Will be closed at the end of context.
         """
         with mlflow.start_run(
             run_name=run_config.name,

src/bikes/utils/signers.py

@@ -21,7 +21,7 @@
 class Signer(abc.ABC, pdt.BaseModel, strict=True, frozen=True, extra="forbid"):
     """Base class for generating model signatures.
 
-    Allow to switch between model signing strategies.
+    Allow switching between model signing strategies.
     e.g., automatic inference, manual model signature, ...
 
     https://mlflow.org/docs/latest/models.html#model-signature-and-input-example

src/bikes/utils/splitters.py

@@ -132,7 +132,9 @@ def split(
         targets: schemas.Targets,
         groups: Index | None = None,
     ) -> TrainTestSplits:
-        splitter = model_selection.TimeSeriesSplit(n_splits=self.n_splits, test_size=self.test_size)
+        splitter = model_selection.TimeSeriesSplit(
+            n_splits=self.n_splits, test_size=self.test_size, gap=self.gap
+        )
         yield from splitter.split(inputs)
 
     @T.override