Tatau
Wrappers for Neural Network Frameworks for TatauCore
Install the tatau
pip package.
$ pip install git+https://github.com/TatauCloud/tatau.git
Prepare Dataset
The dataset samples should be shuffled first. The train and test parts of a dataset should be split on chunks via directories. Tatau platform will split dataset chunks equally between the workers. The chunks should have about the same count of samples. For classification tasks: every chunk should have all class types (Every worker should be able to train on all classes). The count of chunks should be greater than the count of workers, even better if the count of chunks is divided by the count of workers.
Example: Uploaded to IPFS MNIST dataset: train part and test part.