newbie SageMaker with XGBoost

[winanjaya-mtwi]

Hello All,

I am new to AWS Sagemaker, I try to use XGBoost algorithm but it keeps fail, here are what I have done:

• Create a S3 bucket
• Upload the .csv
• Create labeling jobs (completed)
• Create a notebook instance with XGBoost minist example
• Create training job
  • Use Amazon SageMaker built-in Algorithm as Algorithm source
  • Choose XGBoost Algorithm
    • set num_round to 3
    • set objective to reg:liniear
    • set S3 location to s3://mys3/
    • Set S3 output path s3://mys3/output

after waiting couple minutes, it failed with the following error:

ClientError: Blankspace and colon not found in firstline '1 4a4fc709a737ab971e7a1008a72a930c ...' of file 'groundtruth_tqOAx9PEb2RJ3xlAiRN5khotlbq4KUZfZKT_IN2m9d0j2W2h_GbHJxlp5UgJcomc4BqO8qnpPZiDNKcnGqcELSNMzOm6dXRpvzJUOeAFgFOvWfUrym_pI8z35vKYEG.hUdmlDAXjj6M5LVinEg0N8rdMPKBfDnHAmj_9THczi2Y-_etags_do_not_modify.tmp'. ContentType by defaullt is in libsvm. Please ensure the file is in libsvm format.

I need a guidance

mytry.zip

[JavierLopezT]

Hello,

CSVs passed to XGBoost need to be in a specific format:

No header row Outcome variable in the first column, features in the rest of the columns (there's no ability to drop them during the training process) All columns need to be numeric

[dnissani-ias]

I have satisfied all requirements as described above, and am still getting the same error as above. How do you set the content type to csv for XGBoost?

[sandys]

sorry to add a comment here, but is it preferable to use svmlight format (https://scikit-learn.org/stable/modules/generated/sklearn.datasets.dump_svmlight_file.html) versus CSV ?

[mramakrishnan-chwy]

The confusion is because of not very consistent documentation. Yes, the XGBoost supports both CSV and libsvm. For csv, it worked with s3_input object for me. Here is how I defined it,

train_channel = sagemaker.session.s3_input(_s3_inputtrain, content_type ='csv') valid_channel = sagemaker.session.s3_input(_s3_inputvalidation, content_type ='csv')

_s3_inputtrain and _s3_inputvalidation contains the path to my file in s3 buckets

Now, fit the model with these train and validation s3 file inputs

Ref:https://github.com/aws/sagemaker-python-sdk/issues/133

[engineeryashsaxena]

But what if I am doing it via Sagemaker UI ?