def test_RMSprop(self):
updater = opt.RMSprop()
updates = updater(cost, [b])
func = theano.function(inputs=[x, y],
outputs=cost,
updates=updates,
allow_input_downcast=True)
for i in range(500):
func(data[i], lable[i])
self.assertTrue(func(1, 1) < acceptThreshold)
def reset(self):
"""
For sequential layerout network, use append().
To add more layers, the first layer is set with setInput().
Network can do this, because it remember which layer to append to
by using member variable currentLayer.
"""
self.batchsize = 128
self.currentLayer = None
self.inputLayers = []
#layers = []
self.debug = False
self.X = T.tensor4()
self.Y = T.matrix()
self.inputSizeChecker = None
self.outputSizeChecker = None
self.params = []
self.costFunc = cost.CrossEntropy
self.gradientOpt = opt.RMSprop()
self.regulator = reg.Regulator()
#self.regulator = None
self.learner = None
self.predicter = None
self.cost = None # local variable (tensor)
self.nonlinear = None
self.layerCounter = 0
self.modelSavePath = './expdata'
self.modelSavePrefix = 'saved_model_'
self.modelSaveTimeTemplate = '%Y-%m-%d_%H-%M-%S'
self.latestLinkName = 'LAST'
self.modelSaveInterval = 20
self.modelSaveCounter = 0
self.lastSaveAbsolutePath = None