Use IMDB movie reviews to classify the sentiment, either positive or negative
Credit: Deep Learning with Python by Jason Brownlee
Shape of the data: 50000 reviews Unique classes of the predicted variable: [0, 1] Total number of unique words: 88585 Review length: Mean 234.76 +/- 173 words
Parameters: Maximum unique words: 5000 Maximum length of review: 500 words Mumber of vectors for word embedding: 32 50/50 split of training and validation
batch size 128, only 2 epochs is able to reach an accuracy of 87.16%. It only took a minute to run it.
Two epochs reached an accuracy of 88.23%, and it only took a minutes to run it.3 epoches, batch size 64
Much slower
Final accuracy: 87.62%, comparable to previous, but much much much slower, saved
1-d CNN outperform simple MLP slightly and reached an accuracy of 88.23% with only 2 epoches. Training more epoches will likely to improve the accuracy