Google Colaboratory 上带有 GPU 的 XGBoost

Question

I am trying to use XGBoost with GPU on Google Colaboratory. Here is my notebook:

import numpy as np
import os
import xgboost as xgb

train_X = np.random.rand(100,5)
train_Y = np.random.choice(2, 100)

test_X = np.random.rand(10,5)
test_Y = np.random.choice(2, 10)

xg_train = xgb.DMatrix(train_X, label=train_Y)
xg_test = xgb.DMatrix(test_X, label=test_Y)

param = {}
# use softmax multi-class classification
param['objective'] = 'multi:softmax'
# scale weight of positive examples
param['eta'] = 0.1
param['max_depth'] = 6
param['silent'] = 1
param['nthread'] = 4
param['num_class'] = 2

param['gpu_id'] = 0
param['max_bin'] = 16
param['tree_method'] = 'gpu_hist'


# watchlist allows us to monitor the evaluation result on all data in the list 
watchlist = [(xg_train, 'train'), (xg_test, 'test')]
num_round = 5

bst = xgb.train(param, xg_train, num_round, watchlist)

When I run last line:

bst = xgb.train(param, xg_train, num_round, watchlist)

I get "Runtime died, Automatically restarting"

Any ideas how to troubleshoot?

Answer 1

I have got XGBoost running on Colab with GPU support. Download the Linux version from here , and then

!pip uninstall xgboost
!pip install xgboost-0.81-py2.py3-none-manylinux1_x86_64.whl

...is working for me.

Answer 2

Matt Wehman answer works for me. I had some doubts on how to actually place the file xgboost-0.81-py2.py3-none-manylinux1_x86_64.whl inside Colab after download.

The steps are:

1. Download the XGBoost file from the link provided by Matt onto your local machine

2. Upload the file onto the Colab server. this can be done either directly from your computer, or saving to your Google Drive and importing from Drive into Colab. You will need to upload the file each time you start the Colab session (which resets after some time of non-use) and loading from Drive is much faster than from your PC.

   To upload using your PC:

    from google.colab import files files.upload()

   To upload from Google Drive I install and use pyDrive. The process is described here

3. Once the XGBoost file is in the Colab local directory, you can finally run the code

!pip uninstall xgboost
!pip install xgboost-0.81-py2.py3-none-manylinux1_x86_64.whl

Answer 3

At the moment it looks like !pip install -U xgboost works - seems like the version of xgboost installed on colab was old (0.9.0 or something). The xgboost docs also have a link to the nightly builds for xgboost which can be installed like !pip install https://s3-us-west-2.amazonaws.com/xgboost-nightly-builds/xgboost-1.4.0_SNAPSHOT%2B4224c08cacceba3f83f90e387c07aa6205a83bfa-py3-none-manylinux2010_x86_64.whl from a colab jupyter notebook cell. Since it looks like their list of wheels changes sometimes, you might have to search for something like 'xgboost docs wheels' to find the latest location of the wheels if you want to use those.

Answer 4

据我所知，我们无法在 Google Colab 上导入支持 GPU 的 XGBoost，您检查过了吗？