if (i != 0):
# Load pixel values for that row, as well as class
dataTrain[i-1,:] = row[1:785]
classTrain[i-1,:] = int(row[0])
# Normalize data between 0 and 1
dataTrain[i-1,:] /= 255
# Print progress..
if( i % 1000 == 0) : print("{0} training samples loaded...".format(i))
i = i + 1
# Training Vector
print("Encoding Labels in One-Hot...")
classes = utils.oneHotEncode(classTrain, 10)
correctClassif = np.argmax(classes, axis=1)
# --------------------------------- TRAINING NETWORK ---------------------------------- #
for n in range(NUM_EPOCH):
# Resetting the errors to zero
numErr = 0
# Training the network...
for i in range(TRAIN_SAMP):
prediction = ann.trainBP(dataTrain[i,:].reshape((N_PIXELS,1)), classes[i,:].reshape(10,1))
# Read the dataset from file
dataset = np.zeros((214, 9))
classes = np.zeros((214, 1), dtype=np.uint8)
csvfile = open("./datasets/Glass/glass.data", "r")
csvreader = csv.reader(csvfile, delimiter=",")
for row in csvreader:
dataset[int(row[0])-1,:] = row[1:10]
classes[int(row[0])-1,:] = int(row[10])
# Normalize data
dataset = preprocessing.scale(dataset)
# Training Vector
classes = utils.oneHotEncode(classes, 7)
correctClassif = np.argmax(classes, axis=1)+1
# Testing error
numErr = 0
for n in range(NUM_EPOCH):
# Printing progress
if(n % 100 == 0) :
print("Training Epoch {0}".format(n))
print("Average error: ", numErr/TEST_SAMP)
# Resetting the errors to zero
numErr = 0
