Vladimir Tsyshnatiy
Reputation: 1069
NN Batch and Training accuracy
I have NN based on InceptionV3.
Feeding it batches during epochs gives good results like:
loss: 0.3596 - acc: 0.8479 - val_loss: 0.3442 - val_acc: 0.8515
After training was done I have decided to check training accuracy on the whole training dataset.
sklearn.metrics.accuracy_score(labels, np.round(train_predictions))
gave me 0.52520718232044195
How that is possible?
Upvotes: 1
Views: 219
Answers (1)
Vladimir Tsyshnatiy
Reputation: 1069
50% was because data sent to 'predict' was completely different from the one it was trained on because of the bug in preprocessing.
Upvotes: 2
Related Questions
- Does batch size affects on the accuracy of a model?
- How can I use batch size in neural networks
- is there any impact on model accuracy of batch learning while learning with tensorflow cnn
- Dealing with batch size and time step in 1D CNN
- Bigger batch size reduce training time
- Why does different batch-sizes give different accuracy in Keras?
- How to calculate numbers of parameters in CNN model with BatchNormaliztion
- How to check the learning rate with train_on_batch [Keras]
- Keras: batches not completed
- CNN in Keras: unexpected drop of accuracy for small batch sizes