Invalid specification of input data in windows local mode #844
Description
System Information
- Framework (e.g. TensorFlow) / Algorithm (e.g. KMeans): Scikit-Learn
- Framework Version: 0.20.0 (official sagemaker-scikit-learn-container)
- Python Version: 3.6
- CPU or GPU: CPU
- Python SDK Version: 1.26.0
- Are you using a custom image: No
Describe the problem
When I execute fit method in local mode on windows, incorrect training-data file path will be transferred to the docker container.
Cause
I already know that the cause is the difference between windows and linux file separators. Therefore, I specify the cause.
The following code is provided for sagemaker-python-sdk/src/sagemaker/local/image.py.
self.container_dir = container_dir if container_dir else os.path.join('/opt/ml/input/data', channel)When this code is executed on windows, if channel is train, it becomes /opt/ml/input/data\train and can not be recognized by docker.
Minimal repro / logs
- Exact command to reproduce:
sklearn = SKLearn(
entry_point='scikit_learn_iris.py',
train_instance_type="ml.c4.xlarge",
role=role,
sagemaker_session=sagemaker_session,
hyperparameters={'max_leaf_nodes': 30})
sklearn.fit({'train': train_input})