Stanislaw Jastrzebski

Baseline fix

2

Turns out it is important to calculate counts only based on tokens of given passage, rather than estimated on all stories. It is important to produce similar results as in...

Bidirectional bug?

1

I think you should have added before this https://github.com/shyamupa/snli-entailment/blob/master/amodel.py#L101 a reverse lambda (see https://github.com/fchollet/keras/pull/3495/files for correct bidirectional lstm in keras) :)

BN serialization

3

Using theano and blocks from master branches. BN ``population_mean`` and ``population_std`` are not serialized. To reproduce call the following script (based on ``blocks-examples/mnist/__init__.py``) https://gist.github.com/kudkudak/04089a2a3a2442c935d9a26581b869b4: 1. ``python mnist.py --lr=0.1 --bn_momentum=0.1``. This...

e.g. fetching GloVe takes ~3 minutes, which seems way too high 

Otherwise API might be confusing and do unexpected things

Add evaluation methods on property datasets, compare conceptnet and fasttest

1

https://github.com/lucy3/grounding-embeddings