python deep learning

Python: NotMNIST Data Importer and Startup Code:

import numpy as np

import matplotlib.pyplot as plt

import matplotlib.image as mp

def notMNIST_read (number_of_files,file):

c = 10 #classes A-J (10)

m = number_of_files #samples number of files in the small DB

trainX = np.zeros((m,28*28)) #images are 28*28

trainY = np.zeros((m,c))

ix = 0

for (root, dirs, files) in os.walk(file):

for f in files:

if f.endswith('.png'):

try:

img = mp.imread(os.path.join(root,f))

trainX[ix,:]=np.asarray(img).reshape(28*28)

folder = os.path.split(root)[-1]

letter = ord(folder)-ord('A') #A=0, B=1,..., J=9

trainY[ix,letter]=1

print("processing letter %s, file number %d"%(folder, ix),'\r', end="", f

lush=True)

ix+=1

except:

pass

if m==18724:

np.save('notMNIST_small_trainX', (trainX[:ix,:]*255).astype('uint8'))

np.save('notMNIST_small_trainY', (trainY[:ix,:]).astype('uint8'))

elif m==529114:

np.save('notMNIST_Large_trainX', (trainX[:ix,:]*255).astype('uint8'))

np.save('notMNIST_Large_trainY', (trainY[:ix,:]).astype('uint8'))

return trainX, trainY

trainX,trainY = notMNIST_read(18724,"C:/location/small")

trainX = np.load('notMNIST_small_trainX.npy')/255

trainY = np.load('notMNIST_small_trainY.npy')/1

def display_data(ix,X=trainX,Y=trainY):

plt.imshow(X[ix,:].reshape(28,28),cmap='gray')

array=['A','B','C','D','E','F','G','H','I','J']

plt.title(array[np.argmax(Y[ix,:])])

def normalizeX (X):

Xm = np.mean(X,axis=1,keepdims=True)

Xstd = np.std(X,axis=1,keepdims=True)

Xnan = np.where(Xstd==0)

Xstd[Xnan]=1

Xm [Xnan]=0

X -= Xm

X /= Xstd

return X

def trainsplit(X,Y,percentV=0.2, percentT=0.2):

m = X.shape[0]

ix = list(range(m))

np.random.shuffle (ix) #randomize the DB index

X = X[ix,:]

Y = Y[ix,:]

m_train = np.floor(m*(1-percentV-percentT)).astype(int) #cut off for training

m_val = m_train + np.floor(m*percentV).astype(int) #cut off for validation data

X_train = X[ :m_train,:]

X_val = X[m_train:m_val, :]

X_test = X[ m_val:, :]

Y_train = Y[ :m_train,:]

Y_val = Y[m_train:m_val, :]

Y_test = Y[ m_val:, :]

return X_train, Y_train, X_val, Y_val, X_test, Y_test