Still in epoch 1 about cs224n-squad-project

[vincentandreas]

@HarnishLJ The flag is mode=show_examples. (from the previous is mode=train)And this is the result when I execute that code.

!python /content/gdrive/"My Drive"/"Colab Notebooks"/cs224n-Squad-Project-master/code/main.py --experiment_name=aku_nyoba --dropout=0.15 --batch_size=60 --hidden_size_encoder=150 --embedding_size=100 --do_char_embed=False --add_highway_layer=True --rnet_attention=False --bidaf_attention=True --answer_pointer_RNET=False --smart_span=True --hidden_size_modeling=150 --mode=show_examples --gpu=1  --train_dir /content/gdrive/"My Drive"/"Colab Notebooks"/experiments

I'm sorry, I haven't implemented my idea, to change the code, because for now I try to adjust the dataset for my local language.

[vincentandreas]

hey @HarnishLJ, I found how to test using your own testing data, just make a json file, then use official_eval mode.

!python /content/gdrive/"My Drive"/"Colab Notebooks"/cs224n-Squad-Project-master/code/main.py --experiment_name=aku_nyoba --dropout=0.15 --batch_size=60 --hidden_size_encoder=150 --embedding_size=100 --do_char_embed=False --add_highway_layer=True --rnet_attention=False --bidaf_attention=True --answer_pointer_RNET=False --smart_span=True --hidden_size_modeling=150 --mode=official_eval --gpu=1  --train_dir /content/gdrive/"My Drive"/"Colab Notebooks"/experiments   --json_in_path /content/gdrive/"My Drive"/"Colab Notebooks"/cs224n-Squad-Project-master/data/FOR-TEST.json --ckpt_load_dir=/content/gdrive/"My Drive"/"Colab Notebooks"/experiments

then, the code will make prediction.json file, you can open the file to look the generated answer

*FOR-TEST json file format is like train-v1.1.json format, but without answer, and answer_start, you can edit the json

[vincentandreas]

@HarnishLJ The flag is mode=show_examples. (from the previous is mode=train)And this is the result when I execute that code.

!python /content/gdrive/"My Drive"/"Colab Notebooks"/cs224n-Squad-Project-master/code/main.py --experiment_name=aku_nyoba --dropout=0.15 --batch_size=60 --hidden_size_encoder=150 --embedding_size=100 --do_char_embed=False --add_highway_layer=True --rnet_attention=False --bidaf_attention=True --answer_pointer_RNET=False --smart_span=True --hidden_size_modeling=150 --mode=show_examples --gpu=1  --train_dir /content/gdrive/"My Drive"/"Colab Notebooks"/experiments

I'm sorry, I haven't implemented my idea, to change the code, because for now I try to adjust the dataset for my local language.

@HarnishLJ btw, how good your result by testing using show_examples mode? My result is not too good, just 5 correct from 10 questions, I don't know how to increase the result. Maybe anyone who read this can told us how good when you show_example the model?

[HarnishLJ]

The training will run for infinite time, you have to stop your training when model is fully learn.
But I don't know what to do after that when model is fully learn from dataset.

[vincentandreas]

@HarnishLJ but the strange thing is we still in epoch 1, even we already defined how many epoch when we run the main.py . I'm curious too, how we use trained model to do question answering like @priya-dwivedi website https://deeplearninganalytics.org/demos . Maybe anyone who read this issue can give us help?

[HarnishLJ]

@deperfecto Yes, I found this (See the picture) and In that they write (Number of epochs to train 0. 0 means train indefinitely)

[vincentandreas]

@HarnishLJ ooh, I don't realize it have default value in the flags, and I don't change the num_epoch parameter too, thanks for realize me.

[HarnishLJ]

Ok welcome but tell me what to do next , I have trained this model up to 7000 iteration and still running, tell me where to stop training. and next what to do after stoping training. @deperfecto @priya-dwivedi

[vincentandreas]

in flag mode, you can choose to 'show_examples', that will show random question, predicted answer, and the true answer. I have an idea to edit the code, so it can receive question for the user, and then predict the answer, Although it's strange, in the flag mode doesn't have 'test' mode .

[HarnishLJ]

yes. It will be really helpful to me and all, If you share screenshots how to change code or flags, even I don't know which flag I have to change. Please share step by step screenshots.Thanks @deperfecto @priya-dwivedi

[mady143]

Hi @deperfecto @HarnishLJ @priya-dwivedi @yuzehe ,
Actually I am at initial state only at first i run with --mode=train and in code i had given the no of epochs=5 but its is still in epoch 1 and iterations are increases.
at first i run with
1.python code/main.py --experiment_name=bidaf_best --dropout=0.15 --batch_size=60 --hidden_size_encoder=150 --embedding_size=100 --do_char_embed=False --add_highway_layer=True --rnet_attention=False --bidaf_attention=True --answer_pointer_RNET=False --smart_span=True --hidden_size_modeling=150 --mode=train

supose after some x iterations i tried like

2.python code/main.py --experiment_name=bidaf_best --dropout=0.15 --batch_size=60 --hidden_size_encoder=150 --embedding_size=100 --do_char_embed=False --add_highway_layer=True --rnet_attention=False --bidaf_attention=True --answer_pointer_RNET=False --smart_span=True --hidden_size_modeling=150 --mode=shoow_examples

and later what i have to do for testing purpose getting questions and answers

@deperfecto earlier comments you have discussed like test with some json file how can i do that testing process and how good the result be can you explain me clearly

Thank and Regards,
Manikantha Sekhar.