pytensor
pytensor copied to clipboard
pymc-devs
PyTensor allows you to define, optimize, and efficiently evaluate mathematical expressions involving multi-dimensional arrays.
### Description We're missing this simple rewrite: ``` import pytensor.tensor as pt from pytensor.graph import rewrite_graph x = pt.scalar("x") out = pt.log(pt.gamma(x)) new_out = rewrite_graph(out, include=("canonicalize", "stabilize", "specialize")) new_out.dprint() ```...
### Description ```python import numpy as np x1, x2, y = np.random.normal(size=(3, 1000, 1)) assert np.isclose(x1.T @ y + x2.T @ y, (x1 + x2).T @ y) %timeit x1.T @...
The fact that we wrap a graph in OpFromGraph, doesn't mean we also want to wrap it's gradient. It can also lead to troubles as in https://github.com/pymc-devs/pymc/issues/7657 I suggest adding...
## Description We add a rewrite for matrix inversion when the matrix is triangular. We check three conditions: - If there is a tag which is upper/lower triangular - If...
## Description This test is flakey and responsible for a lot of random headaches. It does not appear to be required to fail? The test links to issue #770, but...
## Description The MLX dispatch for `IncSubtensor` was assuming that the indexes would always be integers, but they can actually be either integers or slices. This PR adds logic to...
Cherry picking changes from #1604 It seems that simply improving the caching of individual Ops gives us a lot of speedup when still jitting the whole graph. THIS is still...
This could cause spurious disconnected errors, because the tensorified variable was not in the graph of the cost ---- 📚 Documentation preview 📚: https://pytensor--1685.org.readthedocs.build/en/1685/
added trust_inputs to pytensor functions, added multiple pytensor backends, added block_until_ready to jax calls ## Description This PR adds a benchmarking notebook that benchmarks several algorithms in different backends and...
