NALU

This is an implementation of the paper "Neural Arithmetic Logic Units", Trask et al., ArXiV, 2018

the code implements the MNIST arithmetic task given in the paper

To start training, run -

python NAC.py

To test the trained model run -

python NAC_test.py

To run the baseline GRU, run -

python GRU.py

To test the baseline model run -

python GRU_test.py

The mean absolute error at different extrapolation ranges can be obtained by changing the sequence_length value.

The results(mean absolute error) have been given below :-

Seq_length	GRU	NAC
10	2.79	1.60
100	371.18	6.65
1000	4370.00	24.96