How to test after the training
63 views
Skip to first unread message
yinxiy...@gmail.com
Apr 14, 2020, 3:39:25 AM4/14/20
to Visual Commonsense Reasoning
Hello, sorry to interrupt. Now that the R2C training and validation is complete, what should I test next?
I've run the following two statements.
python train.py -params multiatt/default.json -folder saves/flagship_answer python train.py -params multiatt/default.json -folder saves/flagship_rationale -rationale
Rowan Zellers
Apr 14, 2020, 2:42:55 PM4/14/20
to Visual Commonsense Reasoning
great! Now you can use the evaluation script which will score your results on both question answering, answer justification, and the joint task (Q->AR).
yinxiy...@gmail.com
Apr 15, 2020, 3:20:32 AM4/15/20
to Visual Commonsense Reasoning
I'm sorry, I asked for my friend and didn't get it right at first. She's done performing Q->AR:
"python eval_![]()
q2ar.py -answer_preds saves/flagship_answer/valpreds.npy -rationale_preds saves/flagship_rationale/valpreds.npy"
But now we don't know how to do the next test, what are the specific instructions?
yinxiy...@gmail.com
Apr 15, 2020, 3:21:25 AM4/15/20
to Visual Commonsense Reasoning
![O8)8$]Q@W3OJTIF1NWDCPHD.png](https://groups.google.com/group/visualcommonsense/attach/19cd4331b3c0d/O8)8$%5DQ@W3OJTIF1NWDCPHD.png?part=0.1&view=1){kind=small}
Rowan Zellers
Apr 15, 2020, 10:44:57 AM4/15/20
to Visual Commonsense Reasoning
No worries, so that computed the accuracy - so you're done :) nothing else is needed.
Dustray Yin
Apr 15, 2020, 1:27:02 PM4/15/20
to Visual Commonsense Reasoning
Thank you for your reply.
![7@BR)[CD91ON6)0)WH]11$Q.png](https://groups.google.com/group/visualcommonsense/attach/1bde0845f782d/7@BR)%5BCD91ON6)0)WH%5D11$Q.png?part=0.1&view=1)
The data for the test is calculated in the paper, and there is test.jsonl in the dataset, is this not used?
Reply all
Reply to author
Forward
0 new messages