Machine LearningNLP
Random search over nine hyperparameters for an MNIST CNN and an IMDB BiLSTM.
Script-style notebook with recorded results.
TensorFlowKerasscikit-learn ParameterSampler
Overview
Random search over learning rate, depth, width, batch size, optimizer, activation, dropout, kernel size, and initialisation, with the best configuration retrained for 10 epochs.
Recorded results
- CNN test accuracy
- 0.9904
- MNIST, best sampled configuration
- Source: Hyperparametersearch_CNN_RNN.ipynb, cell 1
- BiLSTM test accuracy
- 0.8470
- IMDB sentiment, best sampled configuration
- Source: Hyperparametersearch_CNN_RNN.ipynb, cell 1