Memory leak in tfp.math.minimize, reproducible from tutorial notebook
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 35/100
- Issue 类型
- 缺陷
- 描述清晰度
- 基本清楚
- 活跃度
- 停滞
- 技术栈
- jupyter-notebook, python, tensorflow
调研方向
从链接的“Multiple changepoint detection and Bayesian model selection”教程笔记本开始，通过执行所有单元格，然后从“Unknown number of states”部分重新执行来复现该问题。比较多次运行 tfp.math.minimize 时的内存使用情况，并验证每次运行增加约 20 MB 的现象不再发生。
由索引模型根据 Issue 内容生成。
描述
Python version: 3.11.6
tensorflow==2.18.0
# via
# -r requirements.in
# tf-keras
tensorflow-probability==0.24.0
# via -r requirements.in
tf-keras==2.18.0
# via -r requirements.in
I am trying to perform an analysis built on the multiple changepoint detection and Bayesian model selection code found in this tutorial notebook: https://www.tensorflow.org/probability/examples/Multiple_changepoint_detection_and_Bayesian_model_selection
In the process of running the analysis for multiple datasets I observed that the memory used increases with each successive analysis, by about 20 MB per run in my code. I confirmed that this occurs with the original tutorial notebook as well to ensure that there is not an error specific to my code. The easiest way to reproduce is to download the notebook, execute all cells, then rerun from the "Unknown number of states" section. I have tried adding tf_keras.backend.clear_session() between each run, deleting all created objects before rerunning, and running garbage collection, but nothing has released the ~20MB of memory that is allocated with each tfp.math.minimize run.
- 主要语言
- Jupyter Notebook
- 星标
- 4.4k
- 派生
- 1.1k
- PR 合并指标
- 30 天内没有已合并 PR
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
tensorflow/probability 的其他 Issue
-
KL divergence between two `TransformedDistribution`s with a shared non-injective bijector (`AbsoluteValue`) returns the base KL可能已有人在做 @Andreas-Hum 于 8 天前认领。 未关闭
难度 2/5 1-3 小时 新手友好度 72/100
tensorflow/probability#2036 ·
-
难度 3/5 1-2 天 新手友好度 68/100
tensorflow/probability#2034 ·
-
难度 5/5 一周以上 新手友好度 35/100
tensorflow/probability#2021 ·
-
难度 4/5 3-5 天 新手友好度 30/100
tensorflow/probability#2015 ·
-
难度 3/5 1-2 天 新手友好度 48/100
tensorflow/probability#2013 ·
查看 tensorflow/probability 的全部 Issue
相似的 Issue
-
[Bug]: DoubleMLRDDData accepts a score_col that is no data column (as in its docstring example); the running variable then becomes a covariate and RDFlex fails可能已有人在做 @vrjdev 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 78/100
DoubleML/doubleml-for-py#455 ·
维护者通常 1 天内回复
-
Install page: spectarf's CUDA engine fails with PyTorch for CUDA 13.0 (nvvmError: ERROR_INVALID_IR)可能已有人在做 @lukepolson 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 79/100
PyTomography/PyTomography#435 ·
维护者通常 1 天内回复
-
beginner friendly
难度 2/5 1-3 小时 新手友好度 67/100
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 82/100
aws-samples/sample-per-model-bedrock#821 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
instadeepai/InstaNovo-FM#50 ·