dalinvip / cnn-lstm-bilstm-deepcnn-clstm-in-pytorch

In PyTorch Learing Neural Networks Likes CNN、BiLSTM
Apache License 2.0
1.25k stars 342 forks source link

Explanation of printout/log #4

Closed wayne-ss closed 5 years ago

wayne-ss commented 5 years ago

Could you briefly explain "Batch", different "loss" and "acc" in the printout?

# more print here
Batch[395/6920] - loss: 0.693503  acc: 56.2500%(9/16)
Batch[400/6920] - loss: 0.729841  acc: 37.5000%(6/16)
Dev  Accuracy:  Evaluation - loss: 0.655496  acc: 62.3853%(544/872) 
Test Accuracy:  Evaluation - loss: 0.660529  acc: 61.3399%(1117/1821) 
The Current Best Dev Accuracy: 64.7936, and Test Accuracy is :62.8226, locate on 10 epoch.

# more print here
Batch[425/6920] - loss: 0.726647  acc: 50.0000%(8/16)
Batch[430/6920] - loss: 0.720713  acc: 43.7500%(7/16)Model_count 40

Specifically, why it stops at Batch[430/6920] in each epoch? And which "Test Accuracy" should I take at then end of 10 epoch?


FYI, here's my config.cfg (I only select settings that are different from default):

[Model]
CNN = False
LSTM = True

[Optimizer]
Adam = False
SGD = True

[Train]
device = 0
cuda = True
epochs = 10
log_interval = 5
dalinvip commented 5 years ago

These are some of the problems with the output format