Learning DL from Deep Learning Tutorials (http://www.deeplearning.net/tutorial/)
Copyright (c) 2008–2013, Theano Development Team All rights reserved.
Python 3.4 Anaconda installed （ with all requirements met for using Theano）
#####Experiments
Using gpu device 0: GeForce GTX 780 Ti
LogisticRegression
Optimization complete with best validation score of 7.145833 %,with test performance 7.354167 %
The code run for 178 epochs, with 20.405116 epochs/sec
The code for file logistic.py ran for 8.7s
MLP
Optimization complete with best validation score of 1.790000 %,with test performance 1.920000 %
The code run for 1000 epochs, with 0.699620 epochs/sec
The code for file mlp.py ran for 1429.3s
Convolutional MLP
Best validation score of 1.010000 % obtained at iteration 11400, with test performance 0.940000 %
The code for file convolutional_mlp.py ran for 259.7s
Denoising Autoencoders
Training epoch 2499, cost 49.4413
The no corruption code for file autoencoders.py ran for 5514.08s
Training epoch 2499, cost 63.2158
The 30% corruption code for file autoencoders.py ran for 5832.02s
Stacked Denoising Autoencoders
Optimization complete with best validation score of 0.015100 %, on iteration 1650000, with test performance 1.400000 %
The training code for file sdA.py ran for 3056.26s
Recurrent Neural Network
learning epoch 79 >> 100% completed in 29.53 (sec) <<
BEST RESULT: epoch 59 valid F1 97.05 best test F1 94.08 with the model rnnslu
Long Short Term Memory
Train 0.0235235235235 Valid 0.171428571429 Test 0.19
The code run for 74 epochs, with 6.824690 sec/epochs
Training took 505.0s