Assumption handler for Eye fails on a Blockwise Eye: unpacks batched output shape

未关闭
#2,686 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@Sahilkumar-r 已经在做这个了。

开始于 2026年11月18日。

  • #2688 来自 @Sahilkumar-r —— 未关闭

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
25/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
停滞
技术栈
python

调研方向

Start with pytensor/assumptions/selection.py and the _eye handler; compare its shape unpacking with the delegation in pytensor/assumptions/blockwise.py. Reproduce the issue with the vectorized example in the report, then check that compiling it no longer logs the rewrite failure. Pull request #2688 is already open against this issue.

由索引模型根据 Issue 内容生成。

描述

assumptions bug vectorization

Compiling a vectorized graph that adds pt.identity_like to a matrix before cholesky and cho_solve logs a rewrite failure from orthogonal_solve_to_transpose_matmul. The Blockwise handler in pytensor/assumptions/blockwise.py hands the batched node to the core Eye handler, _eye. That handler does rows, cols = node.outputs[0].type.shape, which raises ValueError on the 3D output.

import pytensor
import pytensor.tensor as pt


def jittered_cho_solve(F, v):
    L = pt.linalg.cholesky(F + pt.identity_like(F) * 1e-8)
    return pt.linalg.cho_solve((L, True), v)


F = pt.tensor("F", shape=(4, 2, 2))
v = pt.tensor("v", shape=(4, 2))
out = pt.vectorize(jittered_cho_solve, signature="(p,p),(p)->(p)")(F, v)
pytensor.function([F, v], out)  # logs "Rewrite failure due to: orthogonal_solve_to_transpose_matmul"
traceback
Rewrite failure due to: orthogonal_solve_to_transpose_matmul
node: Blockwise{CholeskySolve{lower=True, b_ndim=1, overwrite_b=False}, (m,m),(m)->(m)}(Blockwise{Cholesky{lower=True, overwrite_a=False}, (m,m)->(m,m)}.0, v)
TRACEBACK:
Traceback (most recent call last):
  ...
  File ".../pytensor/assumptions/blockwise.py", line 11, in _blockwise_delegate
    return infer_assumption_for_node(
  File ".../pytensor/assumptions/core.py", line 411, in infer_assumption_for_node
    output_states = fn(key, op, feature, fgraph, node, input_states)
  File ".../pytensor/assumptions/selection.py", line 89, in _eye
    rows, cols = node.outputs[0].type.shape
ValueError: too many values to unpack (expected 2, got 3)
主要语言
Python
星标
654
派生
258
平均合并
3 天 41 分钟
30 天内合并 PR
42

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

pymc-devs/pytensor 的其他 Issue

查看 pymc-devs/pytensor 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。