gitea_admin/sientia-dataops-model-manager
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SIENTIAPDE-1430: Introduce comprehensive integration testing with JSON-based scenarios and detailed README documentation. Enhance training workflow to support advanced model configurations, including polynomial regression with mandatory scaler validation. Ensure robust prediction handling by calculating training predictions (y_train_pred) before denormalization and automatically configuring datetime indices for time-series operations.

Browse Source
This commit is contained in:
Bruno Domingues
2025-12-18 17:05:10 -03:00
parent 6e8f87b2a3
commit 06fd08dc70
18 changed files with 631 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
model_manager/utils/models/train_model_params.py
View File

@@ -248,6 +248,12 @@ class TrainModelParams:
f'degree must be at least 2 for Polynomial Regression, got {self.degree}'
)
if self.model_name == 'Polynomial Regression' and self.scaler_name == 'None':
raise ValueError(
'scaler_name must be set (e.g., "Standard Scaler") for Polynomial Regression '
'to avoid numerical overflow with large feature values'
)
if self.model_name == 'Linear Regression' and self.degree != 1:
raise ValueError(f'degree must be 1 for Linear Regression, got {self.degree}')

2
model_manager/utils/models/train_model_result.py
View File

@@ -25,6 +25,7 @@ class TrainModelResult:
regr (LinearRegressionModel): The trained linear regression model.
scaler_dict (dict): A dictionary containing the scalers used to scale the features and target values.
y_pred (pd.Series | None): The predicted target values for the testing dataset. Default is None.
y_train_pred (pd.Series | None): The predicted target values for the training dataset. Default is None.
mse_val (float | None): The Mean Squared Error (MSE) of the predictions. Default is None.
mae_val (float | None): The Mean Absolute Error (MAE) of the predictions. Default is None.
r2_val (float | None): The R-squared (R²) value of the predictions. Default is None.
@@ -46,6 +47,7 @@ class TrainModelResult:
regr: LinearRegressionModel
scaler_dict: dict
y_pred: pd.Series | None = None
y_train_pred: pd.Series | None = None
mse_val: float | None = None
mae_val: float | None = None
r2_val: float | None = None

8
model_manager/utils/repository/model_repository.py
View File

@@ -284,10 +284,16 @@ class ModelRepository:
self.logger.error(error_msg)
raise ValueError(error_msg)
if data.y_train_pred is None:
error_msg = 'Training predictions (y_train_pred) are None'
self.logger.error(error_msg)
raise ValueError(error_msg)
# Prepare reference data (training set)
reference_data = pd.concat([data.x_train, data.y_train], axis=1)
reference_data = reference_data.rename(columns={data.params.target_variable: 'target'})
reference_data['prediction'] = data.regr.predict(data.x_train)
# Use pre-calculated predictions (calculated before denormalization to avoid overflow)
reference_data['prediction'] = data.y_train_pred
# Prepare current data (test set)
current_data = pd.concat([data.x_test, data.y_test], axis=1)

79
model_manager/utils/repository/training_repository.py
View File

@@ -67,6 +67,11 @@ class TrainingRepository:
Exception: If data loading, preprocessing, or training fails
"""
data = load_data(uploaded_file, params.line_separator, params.decimal_separator)
# Configure datetime index if timestamp column exists
# Required for TimeSeriesDiscontinuityAnalyzer (static window removal)
data = self._configure_datetime_index(data)
process_data = self._init_data_preprocessor(params)
process_data.fit(data)
data_view = process_data.transform(data)
@@ -128,7 +133,9 @@ class TrainingRepository:
TrainModelResult: Updated result with predictions, denormalized data,
and metrics (mse_val, mae_val, r2_val)
"""
# Calculate predictions BEFORE denormalization (important for polynomial models)
y_pred_array = tmr.regr.predict(tmr.x_test)
y_train_pred_array = tmr.regr.predict(tmr.x_train)
if params.use_scaler:
scaler = tmr.process_data.get_scaler()
@@ -142,6 +149,9 @@ class TrainingRepository:
tmr.y_train = scaler.denormalize_single_input(tmr.y_train, params.target_variable)
tmr.y_test = scaler.denormalize_single_input(tmr.y_test, params.target_variable)
y_pred_array = scaler.denormalize_predictions(y_pred_array, params.target_variable)
y_train_pred_array = scaler.denormalize_predictions(
y_train_pred_array, params.target_variable
)
else:
# Fallback for sklearn StandardScaler: only inverse-transform features
feature_cols = getattr(
@@ -158,11 +168,15 @@ class TrainingRepository:
tmr.y_pred = pd.Series(y_pred_array, index=tmr.y_test.index)
tmr.y_pred.name = f'{params.target_variable}_pred'
tmr.y_train_pred = pd.Series(y_train_pred_array, index=tmr.y_train.index)
tmr.y_train_pred.name = f'{params.target_variable}_pred'
tmr.x_train = tmr.x_train.sort_index()
tmr.x_test = tmr.x_test.sort_index()
tmr.y_train = tmr.y_train.sort_index()
tmr.y_test = tmr.y_test.sort_index()
tmr.y_pred = tmr.y_pred.sort_index()
tmr.y_train_pred = tmr.y_train_pred.sort_index()
assert tmr.y_pred is not None, 'y_pred should be set at this point'
@@ -321,3 +335,68 @@ class TrainingRepository:
'interaction_only': params.interaction_only,
'original_features': params.variable_columns,
}
def _configure_datetime_index(self, data: pd.DataFrame) -> pd.DataFrame:
"""
Configure datetime index for the DataFrame.
This method attempts to identify a timestamp column and set it as the
DataFrame index with DatetimeIndex type. This is required for
TimeSeriesDiscontinuityAnalyzer (used in static window removal).
The method looks for common timestamp column names and converts the
first matching column to datetime, then sets it as the index.
Args:
data: Input DataFrame
Returns:
pd.DataFrame: DataFrame with DatetimeIndex if timestamp column found,
otherwise returns original DataFrame unchanged
"""
# If index is already DatetimeIndex, just ensure it's sorted
if isinstance(data.index, pd.DatetimeIndex):
self.logger.info('DataFrame already has DatetimeIndex')
return data.sort_index()
# Common timestamp column names
timestamp_columns = [
'timestamp',
'Timestamp',
'TIMESTAMP',
'date',
'Date',
'DATE',
'datetime',
'DateTime',
]
for col in timestamp_columns:
if col in data.columns:
try:
data[col] = pd.to_datetime(data[col])
data = data.set_index(col)
data = data.sort_index()
self.logger.info(f'Configured datetime index from column: {col}')
return data
except (ValueError, TypeError) as e:
self.logger.warning(f'Failed to convert column {col} to datetime: {e}')
continue
# If no timestamp column found, check if first column looks like a timestamp
first_col = data.columns[0]
try:
# Try to parse first column as datetime
test_values = data[first_col].head(10).dropna()
if len(test_values) > 0:
pd.to_datetime(test_values)
data[first_col] = pd.to_datetime(data[first_col])
data = data.set_index(first_col)
data = data.sort_index()
self.logger.info(f'Configured datetime index from first column: {first_col}')
return data
except (ValueError, TypeError):
pass
self.logger.warning('No timestamp column found - some features may not work correctly')
return data