History

current - 2022-05-30 - 0.00Mb

#64: Fixes bug introduced by recent updates (2022-05-29)
#62: Adds a notebook about convolution (2022-05-27)
#63: Add GraphProto to the documentation (2022-04-28)
#60: Extends notebook coverage (2022-03-07)
#59: Removes unnecessary exceptions (2022-03-06)
#58: Uses new API to retrieve gradient for a model (2022-03-03)
#57: Fix opset for ai.onnx.ml in examples after updating to onnx==1.11 (2022-02-22)
#56: Renames Z into Y_grad in loss functions returning grad (2022-02-14)
#55: Removes unncessary nodes in onnx_derivative (2022-02-14)
#54: Implements a function which returns the gradient (2022-02-13)

#48: Checks gradients are identical to scikit-learn for neural networks (2022-01-25)
#47: Compare gradient values with scikit-learn (2022-01-25)
#45: Adds an example about classification (2022-01-23)
#44: Uses bind_ortvalue_input instead of bind_input to be faster (2022-01-23)
#32: Implements classificiation loss (2022-01-23)
#43: Fixes the training of a binary classifier with weights (2022-01-22)
#42: Implements binary log loss for first API of orttraining (2022-01-22)
#41: Refactors to improve profiling analysis (fix penalty, export onnx graphs) (2022-01-15)
#40: Improves training fwbw with caching. (2022-01-14)
#37: Improves performance of caching (2022-01-09)
#36: Makes sure all plt.show() have been disabled in examples (2022-01-04)
#35: Reduces the number of calls to bind_ortvalue (2022-01-04)

#34: Implements L1, L2 losses for partial training (2022-01-03)
#31: Implements penalty when running the gradient (2022-01-03)
#33: Implements more loss functions (2022-01-02)
#30: Implements different learning rate strategies (2022-01-01)
#26: Implements learning rate from neural network (2022-01-01)
#29: Fixes #16, support weights while training (2021-12-30)
#16: Support weights when training a model (2021-12-30)
#28: Replaces OrtValue by C_OrtValue everywhere (2021-12-19)
#24: Be more consistent with OrtValue, OrtDevice, C and python versions (2021-12-19)
#27: Uses C_OrtDevice everywhere (2021-12-16)
#15: Add example with TrainingAgent (error gradient outside) (2021-12-14)
#25: Implements optimizers with forward, backward functionalities (2021-12-04)
#21: Implements a mechanism that update training weights with SGDRegressor or MLPRegressor (2021-12-04)
#23: Extend documentation (2021-12-01)
#22: Move learning_rate logic in a separate class (2021-12-01)
#20: Experiment markdown rendering (2021-12-01)
#19: Implements training with forward, backward (2021-12-01)
#18: Adds classes to train an ONNX gradient with TrainingAgent (2021-11-27)
#17: Minimize the number of data copy while training a model (2021-11-26)
#14: Optimize DataLoader to use iobinding (avoir copy) (2021-11-26)
#13: Adds function plot_onnxs (2021-11-25)
#12: Add more examples (2021-11-19)

#11: Automates nvprof logs retrieval (2021-10-26)
#10: Add parameter to evaluate the model on test data while training (2021-10-12)
#9: Refactoring documentation (2021-10-08)
#8: Add example to look into neural network on GPU (2021-10-07)
#7: Refactoring (2021-10-04)
#6: Add examples with orttraining (2021-10-04)
#5: Fix examples, update documentation (2021-09-28)
#4: Complex scenarios (2021-07-12)
#3: Replaces OnnxSubOperator by OnnxSubEstimator (2021-03-31)