Machine learning with theano
This project is about a simple experiencie with theano and machine learning. I am using MNIST dataset to generate 35 columns of train (ex.: DNN1.pkl) and then with those params generated by DNN1.pkl I did tests on test_set.
I will show how train the dataset:
Parameters schema: param 0 the program name param 1 define a different call for train or not if 0 execute 35 columns of train if 1 different way param 2 define the numbers of train per column param 3: define the normalization widths, max = 3:9
Examples:
The simple way to train all columns is use:
python mcdnn.py 0 the output will be 35 models generated, probably after days, on /models folder
But, you have a different way to train the mcdnn:
python mcdnn.py 1 5 0 10 12 14 16 18 20 On example above we use arguments like: param 0 = mcdnn.py the programa name param 1 = 1 to define a different call of train method param 2 = 5 to define the numbers of train per column param 3:9 = [0 10 12 14 16 18 20] - the normalization widths