Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

BTW17-1GB with SP #1

Open
PiotrCzapla opened this issue Sep 19, 2018 · 1 comment
Open

BTW17-1GB with SP #1

PiotrCzapla opened this issue Sep 19, 2018 · 1 comment
Assignees

Comments

@PiotrCzapla
Copy link
Member

PiotrCzapla commented Sep 19, 2018

BTW17 DataSet

@PiotrCzapla PiotrCzapla self-assigned this Sep 19, 2018
@PiotrCzapla PiotrCzapla changed the title BTW17-10GB with SP BTW17-1GB with SP Sep 19, 2018
@PiotrCzapla
Copy link
Member Author

First training of the classfier, it seems to overfit but for some reason accuracy goes up

(fastai) pczapla@galatea ~/w/ulmfit4de ❯❯❯ python ./fastai_scripts/train_clas.py --dir-path="work/ge2017" --cuda-id=2 \
    --lm-id="nl-4-ge2017-all" --clas-id="class3-nl-4-ge2017"\
    --bs=40 --cl=12 --lr=0.001 --dropmult 0.5 --sentence-piece-model='sp.model' \
    --nl 4 --use_discriminative True
dir_path work/ge2017; cuda_id 2; lm_id nl-4-ge2017-all; clas_id class3-nl-4-ge2017; bs 40; cl 12; backwards False; dropmult 0.5 unfreeze True startat 0; bpe False; use_clr True;use_regular_schedule False; use_discriminative True; last False;chain_thaw False; from_scratch False; train_file_id
INFO: training set len 20941 divided by 20 is 1, removing that last batch of 1 to avoid exceptions
Epoch:   0%|                                                                              | 0/1 [00:00<?, ?it/sepoch      trn_loss   val_loss   accuracy
    0      0.706689   0.654777   0.728328
Epoch: 100%|█████████████████████████████████████████████████████████████████████| 1/1 [02:26<00:00, 146.29s/it]
Epoch:   0%|                                                                              | 0/1 [00:00<?, ?it/sepoch      trn_loss   val_loss   accuracy
    0      0.655126   0.640256   0.736068
Epoch: 100%|█████████████████████████████████████████████████████████████████████| 1/1 [02:34<00:00, 154.60s/it]
Epoch:   0%|                                                                             | 0/12 [00:00<?, ?it/sepoch      trn_loss   val_loss   accuracy
    0      0.660691   0.600824   0.751161
Epoch:   8%|█████▋                                                              | 1/12 [04:01<44:18, 241.64s/it    1      0.599994   0.569795   0.763545
Epoch:  17%|███████████▎                                                        | 2/12 [07:58<40:01, 240.15s/it    2      0.557705   0.559009   0.775929
Epoch:  25%|█████████████████                                                   | 3/12 [11:54<35:51, 239.03s/it    3      0.529055   0.536978   0.7887
Epoch:  33%|██████████████████████▋                                             | 4/12 [15:51<31:46, 238.29s/it    4      0.456995   0.536071   0.795279
Epoch:  42%|████████████████████████████▎                                       | 5/12 [19:48<27:46, 238.09s/it    5      0.43084    0.538832   0.792183
Epoch:  50%|██████████████████████████████████                                  | 6/12 [23:48<23:51, 238.55s/it    6      0.393469   0.55359    0.789861
Epoch:  58%|███████████████████████████████████████▋                            | 7/12 [27:48<19:55, 239.04s/it    7      0.350108   0.567024   0.790248
Epoch:  67%|█████████████████████████████████████████████▎                      | 8/12 [31:47<15:56, 239.01s/it    8      0.344523   0.563478   0.793344
Epoch:  75%|███████████████████████████████████████████████████                 | 9/12 [35:46<11:57, 239.08s/it    9      0.31005    0.593032   0.794892
Epoch:  83%|███████████████████████████████████████████████████████▊           | 10/12 [39:44<07:57, 238.73s/it    10     0.300432   0.627582   0.793731
Epoch:  92%|█████████████████████████████████████████████████████████████▍     | 11/12 [43:45<03:59, 239.18s/it    11     0.26506    0.632482   0.80031
Epoch: 100%|███████████████████████████████████████████████████████████████████| 12/12 [47:45<00:00, 239.46s/it]

model: fwd_class2-nl-4-ge2017_clas_1.h5 got F1 accuracy on ge17 test1: 0.7579890880748247
confusionmatrix

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

1 participant