A package to systematically train/retrain networks. Uses artemis package, which allows to keep track of all experiments which were ran, then retrieve their results.
- train.py, the main file, you can ran it on AWS in background
- /notebooks, examples of notebook
- /data/data_sources - get raw data from DB to form dataset
- /data/datasets - convert raw proces, volumes etc into matrix datasets
- /models - keras (or may be tensorflow later) models
- create an AWS instance with GPU (p2.xlarge)
- get access to it via ssh: ssh -i "/Users/alex/PycharmProjects/ITT/cert/alex-deeplearn.pem" [email protected]
- git clone https://github.com/IntelligentTrading/ittai.git
- git pull
- add all neccesary training experiments to train.add_all_experiments_variants()
- run train.py in a background: nohup python train.py &
folders: ~/artemis/experiments/
http://artemis-ml.readthedocs.io/en/latest/experiments.html
Experiment fields and methods: https://github.com/QUVA-Lab/artemis/blob/master/artemis/experiments/experiment_record.py