yvchen / jointslu Goto Github PK

haha

Hi,

How do you acquire intent labels for each utterance from the raw ATIS database, i.e. from the directory atis2/, atis3, nov93, dec94? I've reviewed many papers but none of them explicitly mentioned this. I have the raw ATIS data at hand but don't know how to get the intent label annotations. AFAIK, the closest annotation for intents are found in the *.sql files, but that still did not match exactly with what are being used by the community.

Would more than appreciate it if you, or anyone can help on this. Thanks so much!

Hi @yvchen ,
first, thanks for sharing your code!
Where the function history_build() used to run memn2n models can be found?

Thanks
Massimo

I wanted to define a metric for the model. How do I do that from the file in samples?

Where is the data from Microsoft Cortana, from this article

The Graph class, and TimeDistributedDense has been removed from the latest keras version. Please update a code or write at Readme pip freeze important modules
When i check it, i will create a pull request

I noticed some issues with the data split in the ATIS dataset (see visualization of the label distributions here.):

• duplicated data samples - 397 (from 5871)
• no train data for 5 intent and 7 slot labels
• up to 20% of the labels present in the training dataset are not present at all in the dev or test datasets

I believe this could make the interpretation of model performance measures somewhat unreliable, and tried to build an alternative, more balanced
data split here.

Really impressive results, congrats! However, how is the F-measure computed? The scores seem outright incredibly high...

I was wondering which version of keras is used to run this.
For me, I have Keras version of 2.1.2.
Also, I downgrade the keras version to 0.3.3. Then had another importing error.