input_var = T.tensor4('inputs')
target_var = T.ivector('targets')
enable = False
if model_type == 1:
path = 'p1.npz'
filter = [16,32,256]
elif model_type == 2:
path = 'p2.npz'
filter = [8,16,128]
else:
path = 'p3.npz'
filter = [16,32,128]
enable = True
network = CNNL.convnetL(input_var,enable,filter)
if loadmodel:
print ("setting params")
network.setParams(path)
if training:
network.trainNetwork(trainX, trainY, validationX, validationY ,lRate ,num_epochs ,target_var, path)
pred = getPredictions(validationX,input_var)
a = getAcc(pred,validationY)
writeToFile('accuracy.csv',a,"id,accuracy")
if model_type == 3:
pred = getPredictions(tX,input_var)
a = analysis(pred,tY)
rng = numpy.random.RandomState(42)
br = BatchReader2.inputs()
br2 = BatchReader2.inputs(testingData=True)
X, Y = br.getNPArray(2)
testX = br2.getNPArray(2)
print testX.shape
num_epochs = 200
n = X.shape[0]
sizeTrain = 0.8
trainX = X[:n * sizeTrain].reshape(-1, 1, 48, 48)
trainY = Y[:n * sizeTrain].astype('uint8')
validationX = X[n * sizeTrain:].reshape(-1, 1, 48, 48)
validationY = Y[n * sizeTrain:].astype('uint8')
testX = testX.reshape(-1, 1, 48, 48)
input_var = T.tensor4('inputs')
target_var = T.ivector('targets')
network = CNNL.convnetL(input_var)
network.setParams('p3.npz')
#network.trainNetwork(trainX, trainY, validationX, validationY ,0.08 ,num_epochs,target_var,0.0001,'p3.npz')
#network.getParams('p3.npz')
y = np.array([])
for i in range(0, testX.shape[0] / 500):
y = np.append(
y, network.makePred(testX[i * 500:(i + 1) * 500], input_var))
f = file('CNNL_Prediction.csv', 'w')
f.write("Id,Prediction\n")
for i, p in enumerate(y):
f.write("%d,%d\n" % (i + 1, p))
f.close()
if __name__ == '__main__':
rng = numpy.random.RandomState(42)
br = BatchReader2.inputs()
br2 = BatchReader2.inputs(testingData = True)
X, Y = br.getNPArray(2)
testX = br2.getNPArray(2)
print testX.shape
num_epochs = 200
n = X.shape[0]
sizeTrain = 0.8
trainX = X[:n*sizeTrain].reshape(-1,1,48,48)
trainY = Y[:n*sizeTrain].astype('uint8')
validationX = X[n*sizeTrain:].reshape(-1,1,48,48)
validationY = Y[n*sizeTrain:].astype('uint8')
testX = testX.reshape(-1,1,48,48)
input_var = T.tensor4('inputs')
target_var = T.ivector('targets')
network = CNNL.convnetL(input_var)
network.setParams('p3.npz')
#network.trainNetwork(trainX, trainY, validationX, validationY ,0.08 ,num_epochs,target_var,0.0001,'p3.npz')
#network.getParams('p3.npz')
y = np.array([])
for i in range(0,testX.shape[0]/500):
y = np.append(y,network.makePred(testX[i*500:(i+1)*500],input_var))
f = file('CNNL_Prediction.csv','w')
f.write("Id,Prediction\n")
for i,p in enumerate(y):
f.write("%d,%d\n"%(i+1,p))
f.close()