v0.2.2

What's Changed

• Add a Sphinx docs site and an example on every public entrypoint by @jessegrabowski in #112
• Compose transforms on either side of the rule, not just after it by @jessegrabowski in #118
• Export the state helpers a hand-written transform needs by @jessegrabowski in #119
• Write back every statistic the compiled function trains by @jessegrabowski in #136
• Accept shape-derived step counts and bounds by @jessegrabowski in #137
• Add a Sharp Edges page for what surprises users of other frameworks by @jessegrabowski in #138

Full Changelog: v0.2.1...v0.2.2

v0.2.1

Full Changelog: v0.2.0...v0.2.1

v0.2.0

What's Changed

• Let a transformer block initialize its residual projections by @jessegrabowski in #83
• Draw every parameter at construction, and refuse a constant scheme by @jessegrabowski in #84
• Replace the network-wide init scheme with per-parameter initializers by @jessegrabowski in #85
• Share one implementation between adam and adamw by @jessegrabowski in #86
• Stop the frozen-generator check from rejecting inner-graph inputs by @jessegrabowski in #87
• Skip parameters the loss cannot reach by @jessegrabowski in #88
• Serialize initializers, replacing CustomInitializer with a decorator by @jessegrabowski in #89
• Allow a stored learning rate, with one writer per variable by @jessegrabowski in #90
• Add a reduce-on-plateau learning rate policy by @jessegrabowski in #91
• Add a recurrent layer, split into a cell and a scan wrapper by @jessegrabowski in #92
• Hoist a draw out of a scan, so dropout works in a recurrent cell by @jessegrabowski in #93
• Add a GRU cell and layer by @jessegrabowski in #94
• Add an LSTM, and a direction argument with a Bidirectional wrapper by @jessegrabowski in #95
• Add a mask argument, so a padded batch does not disturb the recurrence by @jessegrabowski in #96
• Add a Conv1D layer, lowered to a patch gather and one matmul by @jessegrabowski in #97
• Dispatch convolution to per-backend kernels, and give the core CI no backend by @jessegrabowski in #98
• Add a Conv2D layer, and stop its gradient going through a batched matmul by @jessegrabowski in #99
• Add max and average pooling, dispatched on every backend by @jessegrabowski in #100
• Add Flatten, and take batch norm's statistics per channel by @jessegrabowski in #101
• Make ConvLayerGrad usable on its own as a transposed convolution by @jessegrabowski in #102
• Add zero, constant, reflection and replication padding layers by @jessegrabowski in #103
• Advance every training clock the compiled function reads by @jessegrabowski in #105
• Add skip_if, a step guard with non-finite and large-step conditions by @jessegrabowski in #107
• Compose a rule with trailing transforms, and scale by a schedule by @jessegrabowski in #108
• Unbreak the release build's metadata check by @jessegrabowski in #110
• Bump the publish action past the metadata 2.5 rejection by @jessegrabowski in #111

Full Changelog: v0.1.0...v0.2.0

v0.1.0

What's Changed

• Fix dtype upcast when loading checkpoints by @jessegrabowski in #54
• Add cosine annealing learning rate schedule by @jessegrabowski in #60
• Accept a float, shared variable, or schedule as learning_rate by @jessegrabowski in #61
• Add linear decay learning rate schedule by @jessegrabowski in #62
• Add exponential decay learning rate schedule by @jessegrabowski in #63
• Split the test job into parallel groups and cache compiled graphs by @jessegrabowski in #64
• Add polynomial decay learning rate schedule by @jessegrabowski in #65
• Add step decay learning rate schedule by @jessegrabowski in #66
• Let schedules run in either direction, and compose them by @jessegrabowski in #67
• Add a training clock and advance it once per step by @jessegrabowski in #68
• Allow unknown dimensions in Input by @jessegrabowski in #69
• Let parameters declare their own initializer by @jessegrabowski in #70
• Stop silently training parameters behind a stop-gradient by @jessegrabowski in #71
• Fix state-name collision between adam and adamw by @jessegrabowski in #73
• Name the parameters in the disconnected-input error by @jessegrabowski in #72
• Read schedules off the clock the rule counts on by @jessegrabowski in #74
• Fix RNG seeding and reject generators nothing advances by @jessegrabowski in #75
• Let compile_train take extra updates by @jessegrabowski in #76
• Fold the receptive field into the Xavier fans by @jessegrabowski in #77
• Let every layer take its initializers by @jessegrabowski in #78

Full Changelog: v0.0.7...v0.1.0

v0.0.7

What's Changed

• Cast weights to np.array before checkpointing by @jessegrabowski in #51
• Allow additional outputs in compile_train by @jessegrabowski in #52

Full Changelog: v0.0.6...v0.0.7

v0.0.6

What's Changed

• More optimizers! by @jessegrabowski in #41
• Cleanup and bugfixes by @jessegrabowski in #42
• Bugfixes and cleanup in state.py by @jessegrabowski in #43
• Break import cycles in pytensorf.py and layers.py by @jessegrabowski in #44
• Extract serialization codec out of json_serialize.py by @jessegrabowski in #45
• Split layers.py into a module by @jessegrabowski in #46
• Harmonize BatchNorm layers by @jessegrabowski in #47
• More cleanup by @jessegrabowski in #48
• Memoize optimizer state per rule and parameter by @jessegrabowski in #49
• Fix activation upcasting by @jessegrabowski in #50

Full Changelog: v0.0.5...v0.0.6

v0.0.5

What's Changed

• Refactor optimizers by @jessegrabowski in #31
• Add serialization of network and state by @jessegrabowski in #32
• Add erf-based activations: gelu, swish by @jessegrabowski in #33
• Add LayerNorm by @jessegrabowski in #34
• Add Embedding layer with integer-index lookup by @jessegrabowski in #35
• Mypy cleanup by @jessegrabowski in #36
• Add MultiheadAttention layer by @jessegrabowski in #37
• Update to pytensor>=3.2.3 by @jessegrabowski in #40
• Update repo owner by @jessegrabowski in #39
• Add TransformerBlock and MLP helpers by @jessegrabowski in #38

Full Changelog: v0.0.4...v0.0.5

v0.0.4

What's Changed

• Update imports by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/25
• Switch to hatchling + hatch_vcs build by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/24
• Move parameter extraction logic out of Model class by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/23
• Use shared variables to for model parameters by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/26
• bump pytensor/pymc versions by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/30
• fix: configure PyPI metadata and add python 3.14 classifier by @williambdean in https://github.com/jessegrabowski/pytensor_ml/pull/27

New Contributors

• @williambdean made their first contribution in https://github.com/jessegrabowski/pytensor_ml/pull/27

Full Changelog: jessegrabowski/pytensor_ml@v0.0.3...v0.0.4

v0.0.3

What's Changed

• Add softplus activation function by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/17
• Add BatchNorm2D layer by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/18
• Rename batch norm parameters for clarity by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/19
• Adadelta by @HangenYuu in https://github.com/jessegrabowski/pytensor_ml/pull/9
• AdamW by @HangenYuu in https://github.com/jessegrabowski/pytensor_ml/pull/8

Full Changelog: jessegrabowski/pytensor_ml@v0.0.2...v0.0.3

v0.0.2

What's Changed

• Fix typo in model.initialize_weights by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/15
• Add dropout layer by @jessegrabowski in https://github.com/jessegrabowski/pytensor_ml/pull/16
• Optimizer Unit Testing by @HangenYuu in https://github.com/jessegrabowski/pytensor_ml/pull/10

New Contributors

• @HangenYuu made their first contribution in https://github.com/jessegrabowski/pytensor_ml/pull/10

Full Changelog: jessegrabowski/pytensor_ml@v0.0.1...v0.0.2