which dataset.py is used about attentive-filtering-network HOT 2 CLOSED

Hi, I run the scripts and use the CAttenNet, I get the result as this: Eval eer is always larger than 13% 。
===> Validation set: Average loss: 0.1215 EER: 4.3206
===> evalidation set: Average loss: 0.8836 EER: 15.4397
===> Validation set: Average loss: 0.8836 EER: 15.439
===> Validation set: Average loss: 0.1966 EER: 5.5330
===> evalidation set: Average loss: 1.7949 EER: 21.4176
===> Validation set: Average loss: 0.2288 EER: 4.1896
===> evalidation set: Average loss: 1.9633 EER: 15.7935
===> Validation set: Average loss: 0.1529 EER: 3.8001
===> evalidation set: Average loss: 1.3544 EER: 16.1017
===> Validation set: Average loss: 0.2217 EER: 3.6615
===> evalidation set: Average loss: 1.8525 EER: 14.8484
===> Validation set: Average loss: 0.3128 EER: 3.9870
===> evalidation set: Average loss: 2.3662 EER: 15.8706
===> Validation set: Average loss: 0.2874 EER: 3.8649
===> evalidation set: Average loss: 1.9526 EER: 14.8690
===> Validation set: Average loss: 0.3360 EER: 3.6615
===> evalidation set: Average loss: 2.2905 EER: 14.8313
===> Validation set: Average loss: 0.3263 EER: 3.7478
===> evalidation set: Average loss: 2.2401 EER: 14.6379
===> Validation set: Average loss: 0.3659 EER: 3.5714
===> evalidation set: Average loss: 2.5050 EER: 14.2570
===> Validation set: Average loss: 0.2109 EER: 3.3532
===> evalidation set: Average loss: 1.6002 EER: 13.3984
===> Validation set: Average loss: 0.3679 EER: 3.6608
===> evalidation set: Average loss: 2.5162 EER: 13.8675
===> Validation set: Average loss: 0.2920 EER: 3.5801
===> evalidation set: Average loss: 2.1610 EER: 13.9186
===> Validation set: Average loss: 0.3828 EER: 3.4601
===> evalidation set: Average loss: 2.6925 EER: 14.0448
===> Validation set: Average loss: 0.2259 EER: 3.3510
===> evalidation set: Average loss: 1.7532 EER: 13.4660
===> Validation set: Average loss: 0.3302 EER: 3.3951
===> evalidation set: Average loss: 2.3690 EER: 14.0989
===> Validation set: Average loss: 0.3522 EER: 3.3767
===> evalidation set: Average loss: 2.4965 EER: 13.8957
===> Validation set: Average loss: 0.3221 EER: 3.4581
===> evalidation set: Average loss: 2.3104 EER: 13.9490
===> Validation set: Average loss: 0.3885 EER: 3.5273
===> evalidation set: Average loss: 2.6712 EER: 14.2234
===> Validation set: Average loss: 0.3535 EER: 3.4225
===> evalidation set: Average loss: 2.4902 EER: 13.8658

from attentive-filtering-network.

Hi @Honei
this repo is a bit messy and thus there are many redundant code and models. Try the models in this repo instead: https://github.com/jefflai108/Attentive-Filtering-Network/blob/master/src/attention_neuro/simple_attention_network.py

there are convergence issues (see #2 ). This may have to do with the dataset is small and the model training is thus unstable.

from attentive-filtering-network.