Week 10 - Training LSTM for modeled samples
Proposed net
I have defined an LSTM network to address prediction with modeled images and recurrent networks. The defined network is as follows:
Linear dataset results
- 1 DOF, 8000 training samples, 1000 test samples
- 2 DOF, 8000 training samples, 1000 test samples
Parabolic dataset results
- 1 DOF, 8000 training samples, 1000 test samples
- 2 DOF, 8000 training samples, 1000 test samples
- 3 DOF, 80000 training samples, 10000 test samples
Sinusoidal dataset results
- 1 DOF, 80000 training samples, 10000 test samples
- 2 DOF, 80000 training samples, 10000 test samples
- 3 DOF, 80000 training samples, 10000 test samples
- 4 DOF, 80000 training samples, 10000 test samples
Conclusions
The results improve respect to the non-recurrent structure (MLP). However there is still a prediction limit, this time at 4 DOF sinusoidal dynamic. This shows that recurrence has a great contribution in prediction problems where temporal correlations are important.