I train my network 18 epochs.
And, I eval my model on valid dataset after run one epoch.
However, the results are the same.
** 1 epoch**
[======>. 2442/8000 .] Step: 616ms | Tot: 44m42s
** 1-th valid**
total count: 725
F_measure: [29.56, 17.29, 27.23]
recall: [26.45, 15.42, 24.38]
precision: [36.1, 21.57, 33.3]
** 2 epoch**
[======== 4685/8000 .] Step: 841ms | Tot: 1h27m
** 2-th valid**
total count: 725
F_measure: [30.15, 17.64, 27.92]
recall: [27.62, 16.14, 25.56]
precision: [35.73, 21.21, 33.14]
**3 epoch**
[======== 8000/8000 >] Step: 2s74ms | Tot: 2h29m
epoch: 3, ppl: 0.00037755, time: 8996.131, updates: 8000, accuracy: 47.66
[===>.... 1370/8000 .] Step: 761ms | Tot: 24m51s
**3-th valid**
total count: 725
F_measure: [30.15, 17.64, 27.92]
recall: [27.62, 16.14, 25.56]
precision: [35.73, 21.21, 33.14]
**4 epoch**
[======== 3812/8000 .] Step: 561ms | Tot: 1h10m
**4-th epoch**
total count: 725
F_measure: [30.15, 17.64, 27.92]
recall: [27.62, 16.14, 25.56]
precision: [35.73, 21.21, 33.14]
.
.
.
.
**14 epoch**
[======== 6552/8000 .] Step: 606ms | Tot: 2h2m
epoch: 14, ppl: 0.00029522, time: 8991.541, updates: 32000, accuracy: 55.01
[=>...... 795/8000 ..] Step: 828ms | Tot: 14m39s
**14-th valid**
total count: 725
F_measure: [30.15, 17.64, 27.92]
recall: [27.62, 16.14, 25.56]
precision: [35.73, 21.21, 33.14]
.
.
.
**18 epoch**
[======== 7922/8000 >] Step: 649ms | Tot: 2h28m
epoch: 18, ppl: 0.00028912, time: 9025.469, updates: 40000, accuracy: 55.69
[=====>.. 2165/8000 .] Step: 784ms | Tot: 39m17s
**18-th valid**
total count: 725
F_measure: [30.15, 17.64, 27.92]
recall: [27.62, 16.14, 25.56]
precision: [35.73, 21.21, 33.14]
The results on valid dataset do not change, and the loss on the train dataset decreases.
I hope you give some advice.
Thanks in advance.