def train_classifer_main():
np.random.seed(11)
data_path = '/data3/deepak_interns/vikram/vikram/mnist/'
batch_size = 4
#data_path = '../data/cifar-100-python'
model_save_path = './models'
print 'Loading Dataset'
mnist = mnist_data_set(data_path, batch_size)
print 'done loading'
datasets = mnist.data_sets
print type(datasets['train'].X)
print np.max(datasets['train'].X), np.min(datasets['train'].X)
print np.max(datasets['test'].X), np.min(datasets['test'].X)
print datasets['train'].X.shape, datasets['train'].Y.shape
print datasets['test'].X.shape, datasets['test'].Y.shape
data_shape = datasets['train'].X.shape
data_shape = (batch_size, ) + data_shape[1:]
print 'Data shape:', data_shape
path = os.path.join(model_save_path, 'train1')
print 'loading snn'
f = open(os.path.join(path, 'snn_autonet' + '.save'), 'rb')
snn_loaded_object = cPickle.load(f)
f.close()
print('Done')
path = os.path.join(model_save_path, 'classifer')
if not os.path.exists(path):
os.makedirs(path)
os.makedirs(os.path.join(path, 'snapshots'))
print('creating classifer')
classifier = softmax_classifier()
#print('loading classifier')
#f = open(os.path.join(path, 'softmax_classifer' + '.save'), 'rb')
#classifier=cPickle.load(f)
#f.close()
#print('done')
print('classifier TRAINING ...')
train_classifer(classifier, datasets, path, snn_loaded_object)
print('completed training classifer!')
print('saving classifier...')
f = open(os.path.join(path, classifier.name + '.save'), 'wb')
#theano.misc.pkl_utils.dump()
sys.setrecursionlimit(50000)
cPickle.dump(classifier, f, protocol=cPickle.HIGHEST_PROTOCOL)
f.close()
print('Done')
def train_snn_main():
np.random.seed(64)
data_path = '/data3/deepak_interns/vikram/vikram/mnist/'
batch_size = 4
#data_path = '../data/cifar-100-python'
model_save_path = './models'
print 'Loading Dataset'
mnist = mnist_data_set(data_path, batch_size)
print 'done loading'
datasets = mnist.data_sets
print type(datasets['train'].X)
print np.max(datasets['train'].X), np.min(datasets['train'].X)
print datasets['train'].X.shape
#print datasets['valid'].X.shape
data_shape = datasets['train'].X.shape
data_shape = (batch_size, ) + data_shape[1: ]
print 'Data shape:', data_shape
path = os.path.join(model_save_path, 'train1')
if not os.path.exists(path):
os.makedirs(path)
print 'Creating snn'
# print 'loading snn'
# f = open(os.path.join(path, 'snn_autonet' + '.save'), 'rb')
# network=cPickle.load(f)
network = snn(data_shape)
assert(np.amin(network.full_net_layers[-1].W.eval())>=0)
assert(np.amax(network.full_net_layers[-1].W.eval())<=1)
# print(np.amin(network.all_layers[-1].W.eval()))
#np.random.seed(8)
# path = os.path.join(model_save_path, 'snapshots')
# if not os.path.exists(path):
# os.makedirs(path)
print('SNN TRAINING ...')
train_snn(network, datasets, path)
print('completed training snn !')
print('saving trained snn...')
f = open(os.path.join(path, network.name + '.save'), 'wb')
#theano.misc.pkl_utils.dump()
sys.setrecursionlimit(50000)
cPickle.dump(network, f, protocol = cPickle.HIGHEST_PROTOCOL)
f.close()
print('Done')
path = os.path.join(model_save_path, 'train1')
print 'loading snn'
f = open(os.path.join(path, 'snn_autonet' + '.save'), 'rb')
network=cPickle.load(f)
f.close()
print('Done')
batch_size = 4
#snn_loaded_object.all_layers[-1].stdp_enabled=False
data_path = '/data3/deepak_interns/vikram/vikram/mnist/'
print 'Loading Dataset'
mnist = mnist_data_set(data_path, batch_size)
print 'done loading'
datasets = mnist.data_sets
print type(datasets['train'].X)
print np.max(datasets['train'].X), np.min(datasets['train'].X)
data_shape = datasets['train'].X.shape
data_shape = (batch_size, ) + data_shape[1: ]
print 'Data shape:', data_shape
def dog_output(input_image):
_,channels,height,width=input_image.shape