Closed hadilnc01 closed 7 years ago
You can fix this problem easily by modifying the path from "models//SentenceMatch.sample.best.model" to "models/SentenceMatch.sample.best.model".
(Just replace the double slash into a single slash)
The new version fo tensorflow doesn't support double slash in the path.
You don't have to retrain the model. You model has already been stored safely. Go ahead to use the following command line to test your model:
python BiMPM/src/SentenceMatchDecoder.py --in_path test.tsv --word_vec_path wordvec.txt --mode prediction --model_prefix models/SentenceMatch.sample --out_path test.prediction
Ok I ran the command you shared and I was able to get results but I didn't understand the output. I thought I would get a 0 or 1 output in the case of "prediction" and a probability in the case of "probs". Can you please explain to me the output so I can make sense of it?
In the case of "prediction" I got the below sample:
0 0 hh what is hh ? 0 1 what me worry ? what is worrying ? 1 0 where is armenia ? what is armenia ? 0 0 what is e-hrm ? what is hrm ? 1 1 is/was quora down ? was quora down ?
What is the difference between the values in the first column and the second? Which one is the prediction?
In the case of "probs" I got the below sample:
0 1:0.182387366891 0:0.817612707615 0 1:0.663396537304 0:0.336603462696 1 1:0.211678221822 0:0.788321793079 0 1:0.00674327090383 0:0.993256688118 1 1:0.86587280035 0:0.134127125144
I thought that the one with the highest probability would be put in the first column but that's not the case. If you look at the second row, it has a label 0 (in the first column) even though the label 1 has a higher probability.
Apologies for the constant questions but I don't want to assume wrong information :/
For the "prediction" file, the first column is the ground-truth, and the second column is the predictions. You can refer to https://github.com/zhiguowang/BiMPM/blob/master/src/SentenceMatchTrainer.py#L93
For the "probs" file, the first column is the ground-truth, and the following items are the probs for all possible labels. You can refer to https://github.com/zhiguowang/BiMPM/blob/master/src/SentenceMatchTrainer.py#L98
Ok great! Thank you very much for the explanation! 👍
Hello,
The model was finally done with training on the dev set but when it was about to start with the test set, it gave a really big error:
Is there a way to pick up the process from the test set? or do I have to start with the training all over again? Also, why was there an error?