run_analysis_swap_dropout.py

"""
To use this script:
	> python run_analysis.py fit 
	This will save the output to net-specialists.pickle

	 -OR-

	> python run_analysis.py fit name.pickle
	This will save the output to name.pickle
"""

from helpers import *
from reshape import *

net = NeuralNet(
    layers=[
        ('input', layers.InputLayer),
        ('conv1', Conv2DLayer),
        ('pool1', MaxPool2DLayer),
        ('dropout1', layers.DropoutLayer),
        ('conv2', Conv2DLayer),
        ('pool2', MaxPool2DLayer),
        ('dropout2', layers.DropoutLayer),
        ('conv3', Conv2DLayer),
        ('pool3', MaxPool2DLayer),
        ('dropout3', layers.DropoutLayer),
        ('conv4', Conv2DLayer),
        ('pool4', MaxPool2DLayer),
        ('dropout4', layers.DropoutLayer),
        ('hidden5', layers.DenseLayer),
        ('maxout5', Maxout),
        ('dropout5', layers.DropoutLayer),
        ('hidden6', layers.DenseLayer),
        ('maxout6', Maxout),
        ('dropout6', layers.DropoutLayer),
        ('output', layers.DenseLayer),
        ],

    input_shape=(None, 1, PIXELS, PIXELS),
    conv1_num_filters=208, conv1_filter_size=(6, 6), pool1_ds=(2, 2),
    dropout1_p=0.2,
    conv2_num_filters=384, conv2_filter_size=(4, 4), pool2_ds=(2, 2),
    dropout2_p=0.2,
    conv3_num_filters=512, conv3_filter_size=(3, 3), pool3_ds=(2, 2),
    dropout3_p=0.2,
    conv4_num_filters=896, conv4_filter_size=(3, 3), pool4_ds=(2, 2),
    dropout4_p=0.2,

    hidden5_num_units=8192,
    maxout5_ds=2,
    dropout5_p=0.5,    
    
    hidden6_num_units=8192,
    maxout6_ds=2,
    dropout6_p=0.5,
    
    output_num_units=121, 
    output_nonlinearity=nonlinearities.softmax,

    update_learning_rate=theano.shared(float32(0.03)),
    update_momentum=theano.shared(float32(0.9)),

    regression=False,
    batch_iterator_train=DataAugmentationBatchIterator(batch_size=128),
    batch_iterator_test=BatchIterator(batch_size=128),

    on_epoch_finished=[
        AdjustVariable_custom('update_learning_rate', start=0.03, stop=0.001),
        ],
    max_epochs=70,
    verbose=1,
    eval_size=0.0
    )



def fit(output='net-specialists.pickle', fname_pretrain=None):
    if fname_pretrain:
        print "Using Pretrainig:",fname_pretrain
        with open(fname_pretrain, 'rb') as f:
            net_pretrain = pickle.load(f)
    else:
        net_pretrain = None

    specialists = OrderedDict()

    print "Saving Output to:", output
    print "Running for Pixels = ",PIXELS

    for setting in SPECIALIST_SETTINGS:
        cols = setting['columns']
        #X, y = load2d(cols=cols)
        X, y = load2d()

        model = clone(net)
        if 'kwargs' in setting:
            vars(model).update(setting['kwargs'])

        if net_pretrain is not None:
            print "GONNA LOAD>......"
            model.load_weights_from(net_pretrain)

        print("Training model for columns {} for {} epochs".format(
            cols, model.max_epochs))
        model.fit(X, y)
        specialists[cols] = model

    with open(output, 'wb') as f:
        pickle.dump(specialists, f, -1)


if __name__ == '__main__':
    if len(sys.argv) < 2:
        print(__doc__)
    else:
        func = globals()[sys.argv[1]]
        func(*sys.argv[2:])