Error: could not allocate 0 bytes

[hengzhe-zhang]

When I was using this package, I experienced the following problem. According to my observation, there is still a lot of available memory. Thus, what's the problem?

  File "deepforest/tree/_tree.pyx", line 123, in deepforest.tree._tree.DepthFirstTreeBuilder.build
  File "deepforest/tree/_tree.pyx", line 256, in deepforest.tree._tree.DepthFirstTreeBuilder.build
  File "deepforest/tree/_tree.pyx", line 480, in deepforest.tree._tree.Tree._resize_node_c
  File "deepforest/tree/_utils.pyx", line 34, in deepforest.tree._utils.safe_realloc
MemoryError: could not allocate 0 bytes

[xuyxu]

Hi @zhenlingcn, thanks for reporting! Could you print out the data type and shape of your training data, so that we can reproduce the problem.

[hengzhe-zhang]

Yeah, it's very easy to reproduce this problem. We just need to run the following codes:

import numpy as np
from deepforest import CascadeForestRegressor

c = CascadeForestRegressor(n_jobs=1, verbose=0)
c.fit(np.random.randn(10, 5), np.zeros(10, dtype=np.float32))

[hengzhe-zhang]

By the way, I want to point out that some other normal data has also raised this error.

[xuyxu]

Thanks @zhenlingcn, I can reproduce your problem. I will take a careful look latter.

[xuyxu]

This code snippet runs fine:

import numpy as np
from deepforest import CascadeForestRegressor

c = CascadeForestRegressor(n_jobs=1, verbose=2)
c.fit(np.random.randn(10, 5), np.random.randn(10,))

If you want to use DF21 on the fly, could you check if there is a problem after converting the type of y_train into np.float64.

EDIT: We will check where goes wrong when using the target values of type np.float32. This problem may be also caused by the problematic labels np.zeros(10, dtype=np.float32) for regression.

[xuyxu]

@all-contributors please add @zhenlingcn for bug

[allcontributors]

@xuyxu

I've put up a pull request to add @zhenlingcn! 🎉

[hengzhe-zhang]

I don't believe I can solve the problem by simply modifying the data type. In fact, in the given case, it will still raise an error even if we change the type of the data.

[xuyxu]

I agree. Besides, what is the result using the following command on your target values:

import numpy as np
from sklearn.utils.multiclass import type_of_target

print(type_of_target(y_train))

When y_train is np.zeros(10, dtype=np.float32), the result is binary, which is not compatible with CascadeForestRegressor.

[xuyxu]

Hi @zhenlingcn, I will appreciate it if your could test whether the latest PR #44 raises the same error for your problem. The wheels are available at here.

[hengzhe-zhang]

I don't believe the latest PR is an appropriate solution. For example, the following test case is a very reasonable test case. However, the latest version of Deep Forest will raise an error.

import numpy as np
from deepforest import CascadeForestRegressor

c = CascadeForestRegressor(n_jobs=1, verbose=0)
c.fit(np.random.randn(10, 5), np.array([1, 2, 3, 4, 5]))

ValueError: CascadeForestRegressor is used for univariate or multi-variate regression, but the target values seem not to be one of them.

[xuyxu]

Thanks for your feedback, will take a look at the Cython side when I get a moment.

[hengzhe-zhang]

In fact, this problem has hindered me to conduct several comparative experiments. I hope this problem can be solved as soon as possible.

[xuyxu]

Sorry for your problem. Could you check if using the sklearn backend works?

c = CascadeForestRegressor(backend="sklearn")

EDIT: The sklearn backend is slower, but the performance should be the same as the default custom backend, as guaranteed by our unit tests.

[hengzhe-zhang]

It's great! The sklearn backend seems to work well.

[xuyxu]

Glad to here that 😄

[609347781]

ValueError: CascadeForestRegressor is used for univariate or multi-variate regression, but the target values seem not to be one of them.
您好我也遇到了这个问题，设置backend="sklearn"还是会报错，是不是我这边标签列只有0.1的原因（分类的代码可以正常运行）
以下我的数据读取部分
Data = pd.read_excel(r'D:\秭归-巴东段易发性基础数据\第二次实验\预测数据\整体数据.xlsx')
Feature = Data.iloc[1:578158,4:14].values
Label = Data.iloc[1:578158,1].values
print(Label)
print('数据已读取')
#-标准化处理
StandPFeature = preprocessing.StandardScaler().fit_transform(Feature)

#-------2.构造训练集和测试集------#
xTrain = StandPFeature[0:8660,:] #训练集特征
xTest = StandPFeature[len(xTrain):len(StandPFeature),:] #测试集特征
yTrain = Label[:8660:].ravel()
yTest = Label[len(xTrain):len(StandPFeature):].ravel()
print('训练集已分类完毕')
望解答，非常感谢！！！！！

[609347781]

rf1 = CascadeForestRegressor(backend="sklearn")
rf1.fit(xTrain,yTrain)
pred_value=rf1.predict(xTest)

[xuyxu]

你好，数据集的标签列既然只有0、1取值，为啥要用CascadeForestRegressor? @609347781

[609347781]

我也不想做回归的，但是这边后续用所得的概率是做一个地质方面的图，回归模型出的图好看一点（水paper），就用回归模型了 就如果是分类，把最后分类器的是哪一类的概率输出也行，但是我没找到怎么弄，分类器都封装好了...

…

------------------ 原始邮件 ------------------ 发件人: "Yi-Xuan ***@***.***>; 发送时间: 2022年5月5日(星期四) 晚上8:08 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [LAMDA-NJU/Deep-Forest] Error: could not allocate 0 bytes (#41) 你好，数据集的标签列既然只有0、1取值，为啥要用CascadeForestRegressor? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[xuyxu]

请尝试调用CascadeForestClassifier的predict_proba方法，它会返回每个样本属于正类的概率

[609347781]

用sklearn里的随机森林模型也可以正常运行，之前的图就是用随机森林做的，0-1区间概率用绿红渐变色做一下图

…

------------------ 原始邮件 ------------------ 发件人: "LAMDA-NJU/Deep-Forest" ***@***.***>; 发送时间: 2022年5月5日(星期四) 晚上8:08 ***@***.***>; ***@***.******@***.***>; 主题: Re: [LAMDA-NJU/Deep-Forest] Error: could not allocate 0 bytes (#41) 你好，数据集的标签列既然只有0、1取值，为啥要用CascadeForestRegressor? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[609347781]

thanks！

[GOD-TEN]

作者您好，我也有这个问题。我做的是关联预测的方面，我的数据只有0，1。1代表关联，我想预测关联度，所以是个回归问题，但是也会报CascadeForestRegressor is used for univariate or multi-variate regression, but the target values seem not to be one of them这个错误。我查看了源码，发现会对标签进行判断：type_of_target(y)，结果是'binary'就无法用回归问题。请教一下您这可以解决吗。（正常的随机森林可以做回归预测）

[xuyxu]

作者您好，我也有这个问题。我做的是关联预测的方面，我的数据只有0，1。1代表关联，我想预测关联度，所以是个回归问题，但是也会报CascadeForestRegressor is used for univariate or multi-variate regression, but the target values seem not to be one of them这个错误。我查看了源码，发现会对标签进行判断：type_of_target(y)，结果是'binary'就无法用回归问题。请教一下您这可以解决吗。（正常的随机森林可以做回归预测）

可以先尝试把label改成float数据类型，看看能不能绕过type_of_target的判断。不行的话，直接根据报错的traceback，找到代码源文件，然后把对type_of_target的判断注释掉 @GOD-TEN

[GOD-TEN]

非常感谢您提出的方案。第一种：把label改成float数据类型，不可行。不过第二种是可行的，非常感谢！