sydney222 / lstm-mnist

RNN classifier built with Keras to classify MNIST dataset

lstm-mnist

Image classification using a RNN classifier(LSTM) with Keras.

Requirements

• python3
• tensorflow (>=1.4)
• keras (>=2.1.1)
• numpy

Train and evaluate

The classifier is trained on 55k samples and tested on 10k samples (The default split).

The ANN is made of one LSTM layer with 128 hidden units and one dense output layer of 10 units with softmax activation. The rmsprop optimizer is used with categorial_crossentropy as loss function.

Launch lstm_classifier.py to train and evaluate the classifier, you can dump a trained classifier and load it later.

python lstm_classifier.py

Ressources

• Long Short Term Memory
• Understanding LSTMs
• MNIST Dataset