Compare multiple ML experiments using Dash
pip install .
kahmpehr --logdir <directory with data> --port <port>
2 files are needed in logdir:
- results.csv contains all data to be displayed at kahmpehr. The common approach is to append a new row to this csv each time a ML experiment finishes.
- columns_metadata is a dictionary, saved via joblib, which contains information about columns structure. For example, as the csv may potentially contain many columns, it is nice to have them organized like:
{'metadata': [n_layers, optimizer, batch_size, ...],
'metrics': [accuracy, precision, recall, f1score,...]}
Then, the filter widget will be organized in the same way.
