def prediction_pipeline(model):
r"""AlphaPy Prediction Pipeline
Parameters
----------
model : alphapy.Model
The model object for controlling the pipeline.
Returns
-------
None : None
Notes
-----
The saved model is loaded from disk, and predictions are made
on the new testing data.
"""
logger.info("Predict Mode")
# Unpack the model specifications
directory = model.specs['directory']
drop = model.specs['drop']
extension = model.specs['extension']
feature_selection = model.specs['feature_selection']
model_type = model.specs['model_type']
rfe = model.specs['rfe']
separator = model.specs['separator']
# Get all data. We need original train and test for interactions.
partition = Partition.predict
X_predict, _ = get_data(model, partition)
# Load feature_map
model = load_feature_map(model, directory)
# Log feature statistics
logger.info("Feature Statistics")
logger.info("Number of Prediction Rows : %d", X_predict.shape[0])
logger.info("Number of Prediction Columns : %d", X_predict.shape[1])
# Apply treatments to the feature matrix
all_features = apply_treatments(model, X_predict)
# Drop features
all_features = drop_features(all_features, drop)
# Create initial features
all_features = create_features(model, all_features)
# Generate interactions
all_features = create_interactions(model, all_features)
# Remove low-variance features
all_features = remove_lv_features(model, all_features)
# Load the univariate support vector, if any
if feature_selection:
logger.info("Getting Univariate Support")
try:
support = model.feature_map['uni_support']
all_features = all_features[:, support]
logger.info("New Feature Count : %d", all_features.shape[1])
except:
logger.info("No Univariate Support")
# Load the RFE support vector, if any
if rfe:
logger.info("Getting RFE Support")
try:
support = model.feature_map['rfe_support']
all_features = all_features[:, support]
logger.info("New Feature Count : %d", all_features.shape[1])
except:
logger.info("No RFE Support")
# Load predictor
predictor = load_predictor(directory)
# Make predictions
logger.info("Making Predictions")
tag = 'BEST'
model.preds[(tag, partition)] = predictor.predict(all_features)
if model_type == ModelType.classification:
model.probas[(tag,
partition)] = predictor.predict_proba(all_features)[:, 1]
# Get date stamp to record file creation
d = datetime.now()
f = "%Y%m%d"
timestamp = d.strftime(f)
# Save predictions
save_predictions(model, tag, partition)
def prediction_pipeline(model):
r"""AlphaPy Prediction Pipeline
Parameters
----------
model : alphapy.Model
The model object for controlling the pipeline.
Returns
-------
None : None
Notes
-----
The saved model is loaded from disk, and predictions are made
on the new testing data.
"""
logger.info("Predict Mode")
# Unpack the model specifications
directory = model.specs['directory']
drop = model.specs['drop']
feature_selection = model.specs['feature_selection']
model_type = model.specs['model_type']
rfe = model.specs['rfe']
# Get all data. We need original train and test for encodings.
X_train, y_train = get_data(model, Partition.train)
partition = Partition.predict
X_predict, _ = get_data(model, partition)
# Load feature_map
model = load_feature_map(model, directory)
# Log feature statistics
logger.info("Feature Statistics")
logger.info("Number of Prediction Rows : %d", X_predict.shape[0])
logger.info("Number of Prediction Columns : %d", X_predict.shape[1])
# Apply transforms to the feature matrix
X_all = apply_transforms(model, X_predict)
# Drop features
X_all = drop_features(X_all, drop)
# Create initial features
X_all = create_features(model, X_all, X_train, X_predict, y_train)
# Generate interactions
X_all = create_interactions(model, X_all)
# Remove low-variance features
X_all = remove_lv_features(model, X_all)
# Load the univariate support vector, if any
if feature_selection:
logger.info("Getting Univariate Support")
try:
support = model.feature_map['uni_support']
X_all = X_all[:, support]
logger.info("New Feature Count : %d", X_all.shape[1])
except:
logger.info("No Univariate Support")
# Load the RFE support vector, if any
if rfe:
logger.info("Getting RFE Support")
try:
support = model.feature_map['rfe_support']
X_all = X_all[:, support]
logger.info("New Feature Count : %d", X_all.shape[1])
except:
logger.info("No RFE Support")
# Load predictor
predictor = load_predictor(directory)
# Make predictions
logger.info("Making Predictions")
tag = 'BEST'
model.preds[(tag, partition)] = predictor.predict(X_all)
if model_type == ModelType.classification:
model.probas[(tag, partition)] = predictor.predict_proba(X_all)[:, 1]
# Save predictions
save_predictions(model, tag, partition)
# Return the model
return model