Release Notes

hwm v1.1.4

Released on 2024-04-27

Enhancements

  • Support for Probability Predictions in `twa_score` Metric

    The hwm.metrics.twa_score() metric in hwm.metrics has been enhanced to handle probability predictions. This allows users to pass either class labels or probability distributions as predictions, providing greater flexibility in evaluating classification models.

    Example:

    import numpy as np
from hwm.metrics import twa_score

y_true = np.array([1, 0, 1, 1, 0, 1, 0])
y_pred_proba = np.array([
    [0.43678459, 0.56321541],
    [0.68025367, 0.31974633],
    [0.72183598, 0.27816402],
    [0.36490226, 0.63509774],
    [0.55725326, 0.44274674],
    [0.64382937, 0.35617063],
    [0.54868051, 0.45131949]
])

score = twa_score(y_true, y_pred_proba, alpha=0.8)
print(score)
# Output: 0.7936507936507937

  • Batch Computation in `HammersteinWienerRegressor`

    The HammersteinWienerRegressor in hwm.estimators has been optimized to handle larger datasets by implementing batch computation. This improvement resolves memory errors encountered in version 1.0.1 and enhances the model’s scalability and performance when working with extensive data.

    Example:

    import numpy as np
from hwm.estimators import HWRegressor

# Generate large synthetic dataset
X = np.random.rand(1000000, 10)
y = np.random.rand(1000000)

# Initialize and fit the regressor with batch computation
model = HWRegressor(batch_size=10000)
model.fit(X, y)

# Make predictions
predictions = model.predict(X)
print(predictions[:5])
# Output: [0.523, 0.489, 0.501, 0.478, 0.495]

Enhancements

  • Renamed Classes for Improved Usability

    The class names hwm.estimators.HammersteinWienerRegressor and hwm.estimators.HammersteinWienerClassifier have been renamed to hwm.estimators.HWRegressor and hwm.estimators.HWClassifier respectively for brevity and ease of use.

    Deprecation Notice:

    • The old class names HammersteinWienerRegressor and HammersteinWienerClassifier are now deprecated and will be removed in version 1.1.3. Users are encouraged to transition to the new class names to ensure future compatibility.

    Example:

    from hwm.estimators import HWRegressor, HWClassifier

# Initialize the regressor
regressor = HWRegressor(batch_size=10000)
regressor.fit(X_train, y_train)
predictions = regressor.predict(X_test)

# Initialize the classifier
classifier = HWClassifier(batch_size=10000)
classifier.fit(X_train, y_train)
class_predictions = classifier.predict(X_test)
class_probabilities = classifier.predict_proba(X_test)

    Using Deprecated Class Names:

    import warnings
from hwm.estimators import HammersteinWienerRegressor

# To display deprecation warnings
warnings.simplefilter('default', DeprecationWarning)

# Initialize the deprecated regressor
regressor = HammersteinWienerRegressor(batch_size=5000)
# Output: DeprecationWarning: HammersteinWienerRegressor is deprecated
# and will be removed in version 1.2. Use HWRegressor instead.

Bug Fixes

  • **Resolved Memory Errors in hwm.estimators.HWRegressor **

    Addressed memory consumption issues in the HWRegressor when processing large datasets by introducing batch processing mechanisms. This fix ensures stable and efficient model training and prediction without exhausting system memory.

Documentation Updates

  • Updated the documentation in hwm/doc/source/ to reflect the new capabilities of the twa_score metric, including handling of probability predictions. Users can refer to the updated metrics module documentation for detailed usage instructions and examples.

Upgrade Notes

  • Users upgrading from version 1.0.1 to 1.1.4 should ensure that their workflows accommodate the new batch processing parameters in hwm.estimators.HWRegressor.

  • The hwm.metrics.twa_score() function now accepts both label arrays and probability arrays. Ensure that the input format aligns with the desired usage.

Known Issues

  • No known issues at this time. Future updates will address any emerging bugs or feature requests.

Contributors

  • Thanks to all contributors who reported issues, provided feedback, and contributed code to make this release possible.