首页 > 编程问答 >无法序列化 HighLevelGraph 类型的对象

无法序列化 HighLevelGraph 类型的对象

时间：2024-07-23 07:40:07浏览次数：5  
标签：python dask distributed

我编写的代码如下， 

import dask.array as da
from distributed import Client

remote_server = 'localhost:8786'

client = Client(remote_server)

rows, cols = 4123, 4123
chunk_rows, chunk_cols = 1024, 1024

matrix1 = da.random.random(size=(rows, cols), chunks=(chunk_rows, chunk_cols))
matrix2 = da.random.random(size=(rows, cols), chunks=(chunk_rows, chunk_cols))

print(matrix1.compute_chunk_sizes())
print(matrix1.compute_chunk_sizes())

sz = matrix1.shape
dim0 = sz[0]
dim1 = sz[1]
chunk_remain = dim1 % chunk_rows
lack_rows = chunk_rows - chunk_remain
lack_cols = lack_rows

# Find the missing component where splitting into chunks does not make the chunks square

lack_mat0 = da.zeros(shape=[lack_cols, dim1], chunks=(chunk_rows, chunk_cols))
lack_mat1 = da.zeros(shape=[dim0 + lack_cols, lack_rows], chunks=(chunk_rows, chunk_cols))

# Combine generated components

new_arr0 = da.append(matrix1, lack_mat0, axis=0)
new_arr1 = da.append(new_arr0, lack_mat1, axis=1)

new_matrix1 = new_arr1

sz = matrix2.shape
dim0 = sz[0]
dim1 = sz[1]
chunk_remain = dim0 % chunk_rows
lack_rows = chunk_rows - chunk_remain
lack_cols = lack_rows

# Find the missing component where splitting into chunks does not make the chunks square

lack_mat0 = da.zeros(shape=[lack_cols, dim1], chunks=(chunk_rows, chunk_cols))
lack_mat1 = da.zeros(shape=[dim0 + lack_cols, lack_rows], chunks=(chunk_rows, chunk_cols))

# Combine generated components

new_arr0 = da.append(matrix2, lack_mat0, axis=0)
new_arr1 = da.append(new_arr0, lack_mat1, axis=1)

new_matrix2 = new_arr1

# Unit matrix the added component parts

for i in range(rows, new_matrix1.shape[1]):
    new_matrix1[i, i] = 1.0
    new_matrix2[i, i] = 1.0

# Reorganize chunks

new_new_matrix1 = new_matrix1.rechunk((chunk_rows, chunk_cols))
new_new_matrix2 = new_matrix1.rechunk((chunk_rows, chunk_cols))

result_graph  = da.linalg.solve(new_new_matrix1, new_new_matrix2)

print(result_graph.compute_chunk_sizes())

result_future = client.compute(result_graph)

result_result = client.gather(result_future)

print(result_result)

运行脚本 python solve-test-104.py

我收到错误如下， 

2024-07-22 18:28:03,886 - distributed.protocol.pickle - ERROR - Failed to serialize <ToPickle: HighLevelGraph with 1 layers.
<dask.highlevelgraph.HighLevelGraph object at 0x7aa31b0cdd20>
 0. 134840950195584
>.
Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/pickle.py", line 63, in dumps
    result = pickle.dumps(x, **dump_kwargs)
RecursionError: maximum recursion depth exceeded while pickling an object

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/pickle.py", line 68, in dumps
    pickler.dump(x)
RecursionError: maximum recursion depth exceeded while pickling an object

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/cloudpickle/cloudpickle.py", line 1245, in dump
    return super().dump(obj)
RecursionError: maximum recursion depth exceeded while pickling an object

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/pickle.py", line 81, in dumps
    result = cloudpickle.dumps(x, **dump_kwargs)
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/cloudpickle/cloudpickle.py", line 1479, in dumps
    cp.dump(obj)
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/cloudpickle/cloudpickle.py", line 1249, in dump
    raise pickle.PicklingError(msg) from e
_pickle.PicklingError: Could not pickle object as excessively deep recursion required.
Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/pickle.py", line 63, in dumps
    result = pickle.dumps(x, **dump_kwargs)
RecursionError: maximum recursion depth exceeded while pickling an object

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/pickle.py", line 68, in dumps
    pickler.dump(x)
RecursionError: maximum recursion depth exceeded while pickling an object

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/cloudpickle/cloudpickle.py", line 1245, in dump
    return super().dump(obj)
RecursionError: maximum recursion depth exceeded while pickling an object

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/serialize.py", line 366, in serialize
    header, frames = dumps(x, context=context) if wants_context else dumps(x)
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/serialize.py", line 78, in pickle_dumps
    frames[0] = pickle.dumps(
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/pickle.py", line 81, in dumps
    result = cloudpickle.dumps(x, **dump_kwargs)
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/cloudpickle/cloudpickle.py", line 1479, in dumps
    cp.dump(obj)
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/cloudpickle/cloudpickle.py", line 1249, in dump
    raise pickle.PicklingError(msg) from e
_pickle.PicklingError: Could not pickle object as excessively deep recursion required.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/gorn/Projects/Tohoku-U/solvetest/solve-test-104.py", line 68, in <module>
    print(result_graph.compute_chunk_sizes())
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/dask/array/core.py", line 1500, in compute_chunk_sizes
    tuple(int(chunk) for chunk in chunks) for chunks in compute(tuple(c))[0]
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/dask/base.py", line 662, in compute
    results = schedule(dsk, keys, **kwargs)
  File "/home/gorn/Projects/Tohoku-U/solvetest/venv/lib/python3.10/site-packages/distributed/protocol/serialize.py", line 392, in serialize
    raise TypeError(msg, str_x) from exc
TypeError: ('Could not serialize object of type HighLevelGraph', '<ToPickle: HighLevelGraph with 1 layers.\n<dask.highlevelgraph.HighLevelGraph object at 0x7aa31b0cdd20>\n 0. 134840950195584\n>')

The error message "Could not pickle object as excessively deep recursion required" often arises when dealing with Dask and distributed computing, especially when using client.compute on large Dask arrays or complex Dask graphs. This error signals that the object you are trying to send to the Dask workers (in this case, the result_graph ) is too deeply nested or contains circular references that exceed Python's recursion limit.

Let's break down the potential causes and solutions:

Understanding the Issue:

  1. Dask Graphs and Serialization: When you use client.compute , Dask needs to send the computation graph ( result_graph in your case) to the workers. This graph describes how to compute the result from the input arrays. Dask uses serialization (like pickle or cloudpickle ) to transfer this graph.

  2. Recursion Limit: Python has a built-in recursion limit to prevent stack overflows from infinite recursion. When serializing a deeply nested object, the serialization process itself can hit this limit.

Potential Solutions:

  1. Simplify Your Graph: Complex Dask graphs, particularly those created by iterative or recursive operations, can easily become too deep. Analyze your code:
  2. Avoid Unnecessary Operations: Check for any redundant da.append , rechunk , or array manipulations that might be contributing to graph complexity. Can you achieve the same outcome with fewer steps?

  3. Break It Down: If possible, split your computation into smaller, independent stages. Compute intermediate results, persist them if necessary ( result.persist() ), and then use these persisted results in subsequent stages. This can significantly reduce the graph depth.

  4. Increase Recursion Limit (Use with Caution): You can increase Python's recursion limit using sys.setrecursionlimit(limit) . However, this is generally discouraged as a primary solution. A high recursion limit can lead to stack overflows and crashes if the underlying problem is not addressed.

  5. Use dask.delayed for Custom Functions: If you are applying custom functions within your Dask graph, consider wrapping them with dask.delayed . This can help isolate complex computations and make them more serialization-friendly.

Debugging Tips:

  • Visualize Your Graph: Use result_graph.visualize(filename='graph.pdf') to generate a visual representation of your Dask graph. This can help you identify areas of complexity and potential optimizations.
  • Inspect Chunk Sizes: Ensure your chunk sizes are appropriate for your data and computational resources. Very small chunk sizes can lead to excessively large graphs.

Code Improvements (Without Changing Logic): 

import dask.array as da
from distributed import Client

# ... (Your code for connecting to the client and defining matrix dimensions)

# Function to pad a matrix to a multiple of chunk size
def pad_matrix(matrix, chunk_rows, chunk_cols):
    rows, cols = matrix.shape
    lack_rows = chunk_rows - (rows % chunk_rows) if rows % chunk_rows else 0 
    lack_cols = chunk_cols - (cols % chunk_cols) if cols % chunk_cols else 0
    lack_mat0 = da.zeros(shape=[lack_rows, cols], chunks=(chunk_rows, chunk_cols))
    lack_mat1 = da.zeros(shape=[rows + lack_rows, lack_cols], chunks=(chunk_rows, chunk_cols))
    new_arr = da.append(matrix, lack_mat0, axis=0)
    new_arr = da.append(new_arr, lack_mat1, axis=1)
    return new_arr

# Pad matrices
new_matrix1 = pad_matrix(matrix1, chunk_rows, chunk_cols)
new_matrix2 = pad_matrix(matrix2, chunk_rows, chunk_cols)

# Set diagonal elements (no need to iterate)
new_matrix1 = new_matrix1 + da.diag(da.ones(new_matrix1.shape[0]))
new_matrix2 = new_matrix2 + da.diag(da.ones(new_matrix2.shape[0]))

# Rechunk (consider if truly necessary)
new_matrix1 = new_matrix1.rechunk((chunk_rows, chunk_cols))
new_matrix2 = new_matrix2.rechunk((chunk_rows, chunk_cols))

result_graph = da.linalg.solve(new_matrix1, new_matrix2)

# ... (Compute and gather results)

Remember: The best approach is to prevent excessively complex graphs in the first place. By carefully structuring your Dask computations, you can often avoid serialization issues and improve performance.

标签：python,dask,distributed
From： 78777952

相关文章

  • 这段代码是否保证Python对象被立即删除？
    我正在将Redis异步客户端与Celery一起使用，但在两者之间的集成方面遇到了一些问题。上下文是我需要删除redis.Redis实例（在构造函数中创建）以便关闭连接（该对象有一个close方法，但当asyncio事件循环关闭时我无法使用它，heal_client方法仅在这些情况下才会使用。我的代码如......
  • python selenium 行为错误：AttributeError：“Context”对象没有属性“driver”
    我正在使用pythonselenium与Behavior包一起工作。这是代码：@given('theuserisontheloginpage')defstep_given_user_on_login_page(context):PATH='C:/Users/PycharmProjects/ui_test/chromedriver-win32/chromedriver.exe'context.driver=......
  • python 脚本中的路点用于处理大数据集
    我编写了一个脚本，将一堆来自api的请求写入csv文件。该api中有数千个请求，并且在结束循环/退出程序之前永远不会结束。如何合并航路点，以便如果再次发生停顿，它会自动继续前进并最终打印所有请求？我尝试了一些不起作用的方法，但我不知道下一步该怎么做。以下是使用航路点......
  • Python 中的 SSL 模块不可用（在 OSX 上）
    我在OSX10.13上的virtualenv中运行时遇到问题。我已经运行了pipinstall并且路径brewinstallopenssl指向/usr/local/include/openssl有谁知道如何解决这一问题？在我重新安装../opt/openssl/include/openssl使用python后，这种......
  • AWS Elastic Beanstalk chown PythonPath 错误
    我正在AWS的elasticbeanstalk上部署一个Web应用程序，遇到了同样的错误：[StageApplication].Stoprunningthecommand.Error:chown/var/app/staging/venv/bin/python:nosuchfileordirectory.我在我的环境配置中看到属性：PYTHONPATH：/var/......
  • Python：支持索引的内存对象数据库？
    我正在做一些数据整理，如果我可以将一堆字典放入内存数据库中，然后对其运行简单的查询，这会简单得多。例如，类似：people=db([{"name":"Joe","age":16},{"name":"Jane","favourite_color":"red"},])over_16=db.filter(age__g......
  • 如何构建一维数组的二维数组的特定 Python 结构？
    如何构建一维数组（即行向量）的二维数组的特定结构以满足特定我正在维护的遗留程序的结构？我可以在此结构中生成正确的内容all_measurements[:12]array([[0.,0.,0.,2.],[0.02,0.334,0.04,2.24],[0.04,0.668,0.08,2.48],...........
  • 如何使用 Python Flask 将新的咖啡馆（元素）添加到数据库（SQLite）？
    这是我的代码：@app.route("/add",methods=["POST"])defpost_new_cafe():new_cafe=Cafe(name=request.form.get("name"),map_url=request.form.get("map_url"),img_url=request.form.get("img......
  • 使用 tkinter 为 python 创建 GUI 时如何解决语法错误？
    我是一名Python初学者，决定使用tkinter制作一个小型GUI，该GUI接受用户（潜在餐馆）的3个输入，然后单击按钮后随机输出其中一家餐馆。我不断收到语法错误，并认为它与我的buttonfunc或调用它的命令有关。此代码尚未包含在GUI上输出餐厅的任何位置。任何帮助将不胜感激#Pyth......
  • 在 python 中打开 gnome 终端立即显示为僵尸
    作为背景，我正在编写一个脚本来训练多个pytorch模型。我有一个训练脚本，我希望能够在gnome终端中作为子进程运行。这样做的主要原因是我可以随时关注训练进度。如果我可能有多个GPU，我想在单独的窗口中多次运行我的训练脚本。为了实现这一点，我一直在使用popen。以下代码用于打......

赞助商

阅读排行

网站主页关于我们联系我们网站地图

本网站内容转载自其他媒体，侵权联系[admin##ips99.com]。

Copyright © 2020-2023 IPS99 版权所有 IPS99