rel-2.2.6

What's Changed

• Implement __eq__ for RandomStateType and RandomGeneratorType by @brandonwillard in #640

Full Changelog: rel-2.2.5...rel-2.2.6

rel-2.2.5

What's Changed

• Allow jax_funcify_Join to work without omnistaging by @ricardoV94 in #631
• Convert get_vector_length to a dispatch function by @brandonwillard in #632
• Fix TensorVariable.__getitem__ np.newaxis cases by @brandonwillard in #636

Full Changelog: rel-2.2.4...rel-2.2.5

rel-2.2.4

What's Changed

• Fixed: Inv Numba implementation now correctly returns non-integral in… by @kc611 in #627
• Fixed Dimshuffle for scalar result cases by @kc611 in #625
• Fixed float typecasting for Dot implementation in Numba module by @kc611 in #626
• Fixed Numba Scan failures on multidimensional nit sot outputs by @kc611 in #628
• Add tensor.full_like equivalent to np.full_like by @ricardoV94 in #567

Full Changelog: rel-2.2.3...rel-2.2.4

rel-2.2.3

What's Changed

• Add gammainc(c) derivatives by @ricardoV94 in #513
• Move subtensor rewrites from basic opt to subtensor opt v2 by @spiyer99 in #581
• Fix docstring and type hints for indices_from_subtensor by @brandonwillard in #594
• gitignore stuff produced by tests by @ferrine in #598
• Use alt_prefix field in ScanArgs.get_alt_field by @brandonwillard in #601
• Add an empty_like implementation by @brandonwillard in #602
• Added Numba Scan implementation by @kc611 in #607
• Implement Jax dispatch for Erf Op by @ericmjl in #596
• Make Numba conversion of MakeVector work with mixed input types by @brandonwillard in #610
• Fixed local_useless_switch optimization for boolean conditions by @kc611 in #619
• Added Numba IfElse Op by @kc611 in #615
• avoid appending -m64 to CXXFLAGS for riscv by @alexfanqi in #613

New Contributors

• @spiyer99 made their first contribution in #581
• @ferrine made their first contribution in #598
• @ericmjl made their first contribution in #596
• @alexfanqi made their first contribution in #613

Full Changelog: rel-2.2.2...rel-2.2.3

rel-2.2.2

The following changes were made in this release:

• Removed MKL warning
• Added a function that gets variables in a graph based on debugprint IDs
• aesara.printing.debugprint now handles arbitrary HasInnerGraph Ops
• Added missing HasInnerGraph properties to Scan
• Added a function, aesara.graph.basic.get_var_by_name, that gets variables in a graph by their name or auto_name
• Split aesara.link.numba.dispatch into distinct modules
• Created a HasInnerGraph mixin for Ops with inner graphs
• Added developer conda dependencies to environment.yml
• Added an "auto_name" option to aesara.printing.debugprint
• Moved aesara.debugmode.debugprint to aesara.printing
• Removed ScanMethodsMixin Apply argument handling
• Improved exceptions for the Scan Op
• Removed Scan.*_is_tensor attributes
• Moved Scan printing tests into tests.scan.test_printing
• Inner-graph functions in Scan Ops are now lazily computed
• Removed non-sequence settings from ScanInfo
• Fixed formatting and grammar in scan docstring
• Moved Scan helper methods to ScanMethodsMixin
• Replaced Scan info dict with ScanInfo dataclass
• Made basic aesara.tensor types available at package level (e.g. from aesara.tensor import TensorConstant, TensorVariable)
• Made common aesara.graph objects available at package level (e.g. from aesara.graph import Apply, Variable, Constant, Op, ...)
• Added rewrites to simplify unnecessary logit and sigmoid expressions
• Removed out-of-date conda recipe
• Made Reshape work for Numba scalars
• Added a Numba implementation for BernoulliRV
• The output dtype is now properly set for RandomVariables converted Numba
• Made Clip return Numba scalars
• Fixed in-place updates performed on scalars in Numba
• Numba's Reshape conversion is now guaranteed to get C-ordered arrays
• Added infix dot product to TensorVariables (i.e. A @ b for TensorVariables A and b)

rel-2.2.1

The following changes were made in this release:

• Fixed an issue involving RandomVariable rewrite failures when the same RandomVariables are added as outputs to a FunctionGraph
• Fixed some Numba data type inconsistencies (i.e. some data type values were in string form, and others were unnecessarily converted to Numba types)

rel-2.2.0

The following changes were made in this release:

• TensorConstant is now a subclass of TensorVariable
  • Code that was previously using isinstance(..., TensorVariable) as a means of filtering out TensorConstants will need to be updated
• The following config flags have been removed: gpu_*_elemwise, warn__inc_subtensor1_opt, warn__inc_set_subtensor1, warn__reduce_join, warn__signal_conv2d_interface, warn__vm_gc_bug, warn__gpu_set_subtensor1, warn__subtensor_merge_bug, warn__sum_div_dimshuffle_bug, warn__gpusum_01_011_0111_bug, warn__argmax_pushdown_bug, warn__identify_1pexp_bug.
• Fixed a bug in an aesara.scan warning message
• local_1msigmoid rewrites are applied by default
• PatternSub will no longer return nodes with different output types
• Fixed a bug in PatternSub when get_nodes and values_eq_approx are specified
• Implemented DeepCopyOp.infer_shape
• Added a CholeskySolve Op
• Fixed a scope issue in the C code generated by UltraFastScalarSigmoid

rel-2.1.3

The following changes were made in this release:

• log1p_neg_sigmoid rewrite was moved to the "specialization" pass during optimization
• Added standard_normal to RandomStream
• Fixed OpenMP C code generation for Elemwise
• Improved VM error messages
• Misc. docstring improvements
• Use constant folding to determine RandomVariable broadcast patterns

rel-2.1.2

The following changes were made in this release:

• Added an improved ScanArgs that replaces scan_args

rel-2.1.1

The following changes were made in this release:

• TensorVariable.take and aesara.tensor.subtensor.take now return *Subtensor*s that are later optimized
• Fixed a bug in broadcast_shape_iter when all dims are broadcastable