GloVe、word2vecのファインチューニングを評価する

それぞれの評価データに対する精度を見てみます

glove  評価データに対する精度

Epoch 6/10
4960/4960 [==============================] - 0s 35us/step - loss: 0.1302 - acc: 0.9651 - val_loss: 0.1295 - val_acc: 0.9610
Epoch 7/10
4960/4960 [==============================] - 0s 31us/step - loss: 0.1161 - acc: 0.9692 - val_loss: 0.1237 - val_acc: 0.9628
Epoch 8/10
4960/4960 [==============================] - 0s 35us/step - loss: 0.1038 - acc: 0.9708 - val_loss: 0.1151 - val_acc: 0.9647
Epoch 9/10
4960/4960 [==============================] - 0s 38us/step - loss: 0.1044 - acc: 0.9696 - val_loss: 0.1109 - val_acc: 0.9704
Epoch 10/10
4960/4960 [==============================] - 0s 41us/step - loss: 0.0930 - acc: 0.9732 - val_loss: 0.1070 - val_acc: 0.9685
2126/2126 [==============================] - 0s 29us/step
Test score: 0.107, accuracy: 0.968

word2vec 評価データに対する精度

Epoch 6/10
4960/4960 [==============================] - 0s 68us/step - loss: 0.0417 - acc: 0.9881 - val_loss: 0.0566 - val_acc: 0.9845
Epoch 7/10
4960/4960 [==============================] - 0s 83us/step - loss: 0.0361 - acc: 0.9903 - val_loss: 0.0552 - val_acc: 0.9849
Epoch 8/10
4960/4960 [==============================] - 0s 68us/step - loss: 0.0293 - acc: 0.9915 - val_loss: 0.0564 - val_acc: 0.9849
Epoch 9/10
4960/4960 [==============================] - 0s 68us/step - loss: 0.0265 - acc: 0.9911 - val_loss: 0.0603 - val_acc: 0.9821
Epoch 10/10
4960/4960 [==============================] - 0s 74us/step - loss: 0.0263 - acc: 0.9927 - val_loss: 0.0523 - val_acc: 0.9859
2126/2126 [==============================] - 0s 71us/step
Test score: 0.052, accuracy: 0.986