forecast-tools - v0.4.1

Changed

• All point forecast metrics now validate inputs and raise ValueError or TypeError when y_true and y_pred do not meet specifications
• mean_absolute_percentage_error and symmetric_mean_absolute_percentage_error now raise user warnings when values are close to zero. i.e. inflation of metrics can occur.

Fixed

• PATCH: point forecast metrics can now handle edge case where y_true and y_pred inputs are different data types (e.g. pd.Series and np.array). Prior to v0.4.1 this generated an incorrect metric.
  

- Python
Published by TomMonks 11 months ago

forecast-tools - v0.4.0

Changed

• BREAKING: dropped Python 3.8 support
• BREAKING: Unified interface to winkler_score, absolute_coverage_difference and coverage
• DOCS: upgrade documentation for prediction interval evaluation.
• ENV: pip installs forecast-tools from local python package in edit mode.

Added

• Common validation input parameter function for winkler_score, coverage, and absolute_coverage_difference
• Additional test suite for winkler_score and coverage functions focussing on input parameters.
• metrics.interval_score kitchen sink function will return all interval scores (winkler, acd, coverage) by default or can be limited to selection.
• plotting module: contains the interactive plotly function plot_time_series. Documentation also created.
• ENV: added plotly to package requirements.

Fixed

• PATCH: boot_prediction_intervals percentile estimation using numpy modified to remove deprecated parameter interpolation. Replaced with method parameter.
• PATCH: fixed data.load_emergency_dept date parsing user warning. Added date format parameter to pandas read in.

- Python
Published by TomMonks 11 months ago

forecast-tools - v0.3.0

Changed

• ENV: Development environment upgraded to Python 3.11 and libraries as of July 2024.

Added

• BUILD: pyproject.toml added in migration from setuptools to hatch for package build and test management.

Removed

• BUILD: setup.py, requirements.txt, MANIFEST.in no longer needed with hatch build backend.

Fixed

• TESTING: np.NAN -> np.nan and np.Inf -> np.inf. The original constants have been deprecated in numpy
• MODEL: auto_naive error message patch.

- Python
Published by TomMonks over 1 year ago

forecast-tools - v0.2.1

Changes

• PATCH: fixed typo in setup.py
• DOCS: added missing CITATION.cff

- Python
Published by TomMonks about 2 years ago

forecast-tools - v0.2.0

v0.2.0

• ENV: Updated environment to python 3.11 and package versions to latest in Dec 2023
• PATCH: np.float and np.int were deprecated in numpy v1.20. Updated baseline module to use native 'float' and 'int' instead.
• PATCH: cross_validation_folds now returns a list of tuples instead of jagged array.
• ENHANCEMENT: added metrics.winkler_score and metrics.absolute_coverage_difference
• DOCS: updated documentation to include "coverage" metrics.

- Python
Published by TomMonks about 2 years ago

forecast-tools - v0.1.7

• A more informative error message when training data is too short for naive methods.
• Minor fixes of outdated docstrings in forecast_tools.baseline
• Improvements in the handling of invalid parameters by auto_naive.

- Python
Published by TomMonks about 4 years ago

forecast-tools - v.0.1.6

Fixes:

• setup.py fixed: LICENCE file now included on PyPi deployment

Features:

• forecast_tools.feature_engineering module introduced
• forecast_tools.feature_engineering.sliding_window()

- Python
Published by TomMonks over 4 years ago

forecast-tools - v0.1.5

v0.1.5

Features:

• auto_naive function
• scaled error cross validation
• return raw prediction and test sets from cross validation

bug fixes

• mase no longer raise exception when scaling by Naive forecast 1.

- Python
Published by TomMonks over 5 years ago

forecast-tools - v0.1.4

v0.1.4 Minor release

Documentation updates:

All example notebooks moved to examples/ forecasttools added forecastdev environment so that notebooks work in dev environment. updated docstrings for all naive predict() methods

Minor functionality changes

new fitpredictfunction and additional unit tests

Bug fix

updated setup file so that forecast_tools/data * is included in pypi release.

- Python
Published by TomMonks over 5 years ago

forecast-tools - v0.1.3

Big Updates:

forecasttools.modelselection

Cross validation functions. In particular rolling forecast origin, sliding window and crossvalidationscore Unit tests in tests/testmodelselection.py

forecast_tools.metrics

Mean absolute scaled error. An initial implementation of it. This doesn't yet work with the cross validation code due to the difference in function signature. Propose best way around it is to create a decorator function. Unit tests in test/test_metrics.py

Minor

improved docstrings and updated cross_validation notebook.

- Python
Published by TomMonks over 5 years ago

forecast-tools - v0.1.2

Introduction of unit tests (via pytest) for metrics and baseline modules

Minor bug fixes

Naive methods now work with numpy input as well as pd.DataFrame and pd.Series Naive methods now fail gracefully and controlled if abnormal input included Naive methods prediction intervals benchmarked against R

- Python
Published by TomMonks almost 6 years ago

forecast-tools - v0.1.1

Release forecast_tools v0.1.1

baseline forecast methods: average, drift, seasonal naive, naive 1 and ensemble error metrics for point forecasts and prediction interval coverage toy training/test data documentation / tutorial notebooks limited unit testing

- Python
Published by TomMonks almost 6 years ago

forecast-tools - v0.1.0

Release forecast_tools v0.1.0

baseline forecast methods: average, drift, seasonal naive, naive 1 and ensemble error metrics for point forecasts and prediction interval coverage toy training/test data documentation / tutorial notebooks limited unit testing

- Python
Published by TomMonks almost 6 years ago