The purpose of the application is to predict negative music reviews. There are two python scripts: one for nn training and one for running actual predictions. The training and tests sets are based on 6000 music reviews from the website https://www.darkplanet.pl and their respective scores. Reviews are scored from 1 to 5 where score =<3 is considered negative. A vectorized layer is adapted to take standardized strings of given dictionary size. The Best results were obtained with 30 epochs and dropout 0.2. Training data set is split to validation set with default ratio 0.3.
- Unzip training set and test set:
tar -xzf data.tgz- Train model on set
./train.py- Evaluate predictions
./predict.py data/test/pos/38896.txt
./predict.py data/test/neg/52449.txt