History

current - 2022-02-11 - 0.00Mb

  • issue53: Implements get_trained_onnx to retrieve the trained model (2022-02-04)

  • issue51: Implements scoring functions (2022-02-02)

  • issue49: Check nan values during training (2022-01-30)

0.4.274 - 2022-01-29 - 0.06Mb

  • issue48: Checks gradients are identical to scikit-learn for neural networks (2022-01-25)

  • issue47: Compare gradient values with scikit-learn (2022-01-25)

  • issue45: Adds an example about classification (2022-01-23)

  • issue44: Uses bind_ortvalue_input instead of bind_input to be faster (2022-01-23)

  • issue32: Implements classificiation loss (2022-01-23)

  • issue43: Fixes the training of a binary classifier with weights (2022-01-22)

  • issue42: Implements binary log loss for first API of orttraining (2022-01-22)

  • issue41: Refactors to improve profiling analysis (fix penalty, export onnx graphs) (2022-01-15)

  • issue40: Improves training fwbw with caching. (2022-01-14)

  • issue37: Improves performance of caching (2022-01-09)

  • issue36: Makes sure all plt.show() have been disabled in examples (2022-01-04)

  • issue35: Reduces the number of calls to bind_ortvalue (2022-01-04)

0.3.245 - 2022-01-03 - 0.06Mb

  • issue34: Implements L1, L2 losses for partial training (2022-01-03)

  • issue31: Implements penalty when running the gradient (2022-01-03)

  • issue33: Implements more loss functions (2022-01-02)

  • issue30: Implements different learning rate strategies (2022-01-01)

  • issue26: Implements learning rate from neural network (2022-01-01)

  • issue29: Fixes #16, support weights while training (2021-12-30)

  • issue16: Support weights when training a model (2021-12-30)

  • issue28: Replaces OrtValue by C_OrtValue everywhere (2021-12-19)

  • issue24: Be more consistent with OrtValue, OrtDevice, C and python versions (2021-12-19)

  • issue27: Uses C_OrtDevice everywhere (2021-12-16)

  • issue15: Add example with TrainingAgent (error gradient outside) (2021-12-14)

  • issue25: Implements optimizers with forward, backward functionalities (2021-12-04)

  • issue21: Implements a mechanism that update training weights with SGDRegressor or MLPRegressor (2021-12-04)

  • issue23: Extend documentation (2021-12-01)

  • issue22: Move learning_rate logic in a separate class (2021-12-01)

  • issue20: Experiment markdown rendering (2021-12-01)

  • issue19: Implements training with forward, backward (2021-12-01)

  • issue18: Adds classes to train an ONNX gradient with TrainingAgent (2021-11-27)

  • issue17: Minimize the number of data copy while training a model (2021-11-26)

  • issue14: Optimize DataLoader to use iobinding (avoir copy) (2021-11-26)

  • issue13: Adds function plot_onnxs (2021-11-25)

  • issue12: Add more examples (2021-11-19)

0.2.122 - 2021-10-31 - 0.03Mb

0.2.117 - 2021-10-26 - 0.03Mb

  • issue11: Automates nvprof logs retrieval (2021-10-26)

  • issue10: Add parameter to evaluate the model on test data while training (2021-10-12)

  • issue9: Refactoring documentation (2021-10-08)

  • issue8: Add example to look into neural network on GPU (2021-10-07)

  • issue7: Refactoring (2021-10-04)

  • issue6: Add examples with orttraining (2021-10-04)

  • issue5: Fix examples, update documentation (2021-09-28)

  • issue4: Complex scenarios (2021-07-12)

  • issue3: Replaces OnnxSubOperator by OnnxSubEstimator (2021-03-31)

0.1.0 - 2020-07-09 - 0.04Mb

  • issue1: Add an example on black list, while list of operators. (2020-07-09)