pytensor
pytensor copied to clipboard
pymc-devs
PyTensor allows you to define, optimize, and efficiently evaluate mathematical expressions involving multi-dimensional arrays.
### Description - [ ] Implement some basic docs for scalar - [ ] Include link/compile inside compile section - [ ] Move d3viz and include ipython inside printing section...
### Before ```python index = array(['2017-06-02T00:00:00.000000000', '2017-06-03T00:00:00.000000000', '2017-06-04T00:00:00.000000000', ..., '2021-06-28T00:00:00.000000000', '2021-06-29T00:00:00.000000000', '2021-06-30T00:00:00.000000000'], dtype='datetime64[ns]') pt.as_tensor(index) --------------------------------------------------------------------------- KeyError Traceback (most recent call last) File .venv/lib/python3.10/site-packages/pytensor/tensor/type.py:287, in TensorType.dtype_specs(self) 286 try: --> 287 return...
### Description I would be nice for debugging/benchmarking to be allowed to do `function(inputs, outputs, mode='python')`. This would use the `PerformLinker()` with default rewrites (excluding things like `blas_opt` that introduces...
### Description jax mode is missing a dispatch for the `NonZero` `Op`. There's a `jnp.nonzero`, so it should be easy to do.
### Description This make working with these Ops much harder because we always have to check whether the axis is constant (and we can reason about the operation) or not...
### Description This make it hard to import even basic stuff, since some modules want to import stuff defined in the `__init__`, while `__init__` is importing everything else.
### Description Checking https://docs.xarray.dev/en/stable/api.html#top-level-functions The following (unchecked) top level functions are missing. I filtered out ones that are not relevant for PyTensor - [x] broadcast - [x] concat - [...
### Description As the title suggests, graphs with `IfElse` will fail at runtime if you try to vectorize them. The error raised reminds me of #1425 ```py import pytensor.tensor as...
`numba_funcify_Elemwise` can end up with `parent_node` in its kwargs when vectorizing `OpFromGraph`
### Description Trying to compile a graph vectorized with `vectorize_graph` that contains an `OpFromGraph` into numba mode results in an error: ```py X = pt.dmatrix("X", shape=(None, None)) X_batched = pt.tensor("X",...
### Description When compiling a vectorized graph with an `OpFromGraph` into numba mode, shape inference fails if the shape of the output depends on blockwise operations. Example: ```py X =...