Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update abstract PyFunc to utilise returned treatment_config #164

Merged
merged 4 commits into from
Feb 11, 2022
Merged

Update abstract PyFunc to utilise returned treatment_config #164

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f49e303
Update abstract PyFunc to utilise returned treatment_config
deadlycoconuts Feb 8, 2022
e388de3
Rename existing helper functions that to use prefix instead of header
deadlycoconuts Feb 10, 2022
3ecd52e
Update docstrings and duck typing to reflect newly utilised treatment…
deadlycoconuts Feb 10, 2022
2e20c65
Edit docstring description to make it more representative of expected…
deadlycoconuts Feb 11, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 19 additions & 12 deletions sdk/turing/ensembler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ def ensemble(
self,
features: pandas.Series,
predictions: pandas.Series,
treatment_config: Optional[dict]) -> Any:
treatment_config: Optional[pandas.Series]) -> Any:
"""
Ensembler should have an ensemble method, that implements the logic on how to
ensemble final prediction results from individual model predictions and a treatment
Expand All @@ -26,9 +26,9 @@ def ensemble(
:param features: pandas.Series, containing a single row with input features
:param predictions: pandas.Series, containing a single row with all models predictions
`predictions['model-a']` will contain prediction results from the model-a
:param treatment_config: dictionary, representing the configuration of a treatment,
that should be applied to a given record. If the experiment engine is not configured
for this Batch experiment, then `treatment_config` will be `None`
:param treatment_config: Optional[pandas.Series], representing the configuration of a
treatment, that should be applied to a given record/payload. If the experiment
engine is not configured, then `treatment_config` will be `None`

:returns ensembling result (one of str, int, float, double or array)
"""
Expand All @@ -40,8 +40,8 @@ class PyFunc(EnsemblerBase, mlflow.pyfunc.PythonModel, abc.ABC):
Abstract implementation of PyFunc Ensembler.
It leverages the contract of mlflow's PythonModel and implements its `predict` method.
"""

PREDICTION_COLUMN_PREFIX = '__predictions__'
TREATMENT_CONFIG_COLUMN_PREFIX = '__treatment_config__'

def load_context(self, context):
self.initialize(context.artifacts)
Expand All @@ -57,20 +57,27 @@ def initialize(self, artifacts: dict):

def predict(self, context, model_input: pandas.DataFrame) -> \
Union[numpy.ndarray, pandas.Series, pandas.DataFrame]:
prediction_columns = {
col: col[len(PyFunc.PREDICTION_COLUMN_PREFIX):]
for col in model_input.columns if col.startswith(PyFunc.PREDICTION_COLUMN_PREFIX)
}
prediction_columns = PyFunc._get_columns_with_prefix(model_input, PyFunc.PREDICTION_COLUMN_PREFIX)
treatment_config_columns = PyFunc._get_columns_with_prefix(model_input, PyFunc.TREATMENT_CONFIG_COLUMN_PREFIX)

return model_input \
.rename(columns=prediction_columns) \
.rename(columns=treatment_config_columns) \
.apply(lambda row:
self.ensemble(
features=row.drop(prediction_columns.values()),
features=row.drop(prediction_columns.values()).drop(treatment_config_columns.values()),
predictions=row[prediction_columns.values()],
treatment_config=None
treatment_config=row[treatment_config_columns.values()]
), axis=1, result_type='expand')

@staticmethod
def _get_columns_with_prefix(df: pandas.DataFrame, prefix: str):
selected_columns = {
col: col[len(prefix):]
for col in df.columns if col.startswith(prefix)
}
return selected_columns


@ApiObjectSpec(turing.generated.models.Ensembler)
class Ensembler(ApiObject):
Expand Down Expand Up @@ -273,7 +280,7 @@ def create(
conda_env: Union[str, Dict[str, Any]],
code_dir: Optional[List[str]] = None,
artifacts: Dict[str, str] = None,
) -> 'PyFuncEnsembler':
) -> 'PyFuncEnsembler':
"""
Save new pyfunc ensembler in the active project

Expand Down