After each experiment run, dump information about the network in e.g. CSV or JSON

bernhard2202 / twitter-sentiment-analysis

Twitter Sentiment Analysis with Deep Convolutional Neural Networks and LSTMs in TensorFlow.

38 stars 9 forks source link

After each experiment run, dump information about the network in e.g. CSV or JSON #1

Closed AndreiBarsan closed 8 years ago

AndreiBarsan commented 8 years ago

This can let us have a rich log with every single experiment configuration we ran.

We also want to dump this when TF gets C-c'd, so a special Python handler needs to be put into place.

Example:

2016-06-10T22:40, amazon, gpu, $timestamp, $train_loss, $train_acc, $valid_loss, $valid_acc, $tf params, $git_hash ...

AndreiBarsan commented 8 years ago

This seems to be low priority, since we still have the regular TF output with all information neatly stored (this is the default behavior when running stuff on Euler). As long as we run every train operation through tee and dump stdout to some file for later inspection, this feature isn't very useful.

AndreiBarsan commented 8 years ago

I sort of added this when generating the meta.txt file in a run folder, so this issue is no longer really necessary.