Using CCN to train digit recognition

Convolution neural network is known as the most accurate model to train the digit recognition, When in college , we learned laplace transformation in convolution therem . Well I think everyone has return that knowledge to your teacher and leave nothing in your brain. That's ok . So am I. However, we don't need to implement the CNN from stratch and train the data. The keras has already done the model so that we can easily use to trigger the traing in just a few steps.

1. Import Data

import keras
import pandas as pd
from keras import backend as K
from keras.models import Sequential
from keras.layers import Dense, Dropout,Flatten
from keras.layers import Conv2D, MaxPooling2D
from sklearn.model_selection import train_test_split

# Import Data
dataset = pd.read_csv("train.csv")
target = dataset.iloc[:,0].values.ravel()
train = dataset.iloc[:,1:].values
test = pd.read_csv("test.csv").values.reshape(-1,28,28,1)

data_x = train # feartures
data_y = target # targets

x_train, x_test, y_train, y_test = train_test_split(data_x, data_y, test_size=0.2, random_state=42) # split the data
print (x_train.shape);

(33600, 784)

2.Reshape Data

Reshaping data is important as the CNN in Keras only can take data with certain shape. So we have no chices but to reshape the input data.

# 10 numbers 

num_classes = 10  

# input image dimensions
img_rows, img_cols = 28, 28

if K.image_data_format() == 'channels_first':
    x_train = x_train.reshape(x_train.shape[0],1,img_rows,img_cols)
    x_test = x_test.reshape(x_test.shape[0],1,img_rows,img_cols)
    input_shape = (1,img_rows,img_cols)  # 3 = RGB， 1 = black & white
    print (input_shape)
else:
    x_train = x_train.reshape(x_train.shape[0],img_rows,img_cols,1)
    x_test = x_test.reshape(x_test.shape[0],img_rows,img_cols,1)
    input_shape = (img_rows,img_cols,1)
    print (input_shape)


(28, 28, 1)


3. Transfer the integer into 0~1 decimal numbers






# 0~1 float

x_train = x_train.astype("float32")
x_test  = x_test.astype("float32")
x_train /= 255
x_test  /=255 

print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')





4. Set some basic params for the CNN

y_train = keras.utils.to_categorical(y_train,num_classes) # one-hot encoder
y_test = keras.utils.to_categorical(y_test,num_classes) # one-hot encoder

model =Sequential()
model.add(Conv2D(32, kernel_size=(3,3), activation='relu',input_shape=input_shape)) #32C3
          
model.add(Conv2D(64, kernel_size=(3,3), activation='relu')) #64C3

model.add(MaxPooling2D(pool_size=(2,2))) #MP2
model.add(Dropout(0.25))

model.add(Flatten())
model.add(Dense(128,activation='relu')) # Dense must has Flatten in front ,                                 Total Chain : 6C5-MP2-16C5-MP2-120C1 (FLATTEN) -84N-10N

model.add(Dropout(0.5))
model.add(Dense(num_classes,activation='softmax'))

model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])







5. Train the model



The batch_size and epoch are needed to be tuned, here I just randomly give 2 number for demostration.

history = model.fit(x_train, y_train, batch_size=200, epochs=12, verbose=1, validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)



6. Draw a graph

%matplotlib inline

import matplotlib
import matplotlib.pyplot as plt


plt.xlabel('x')
plt.ylabel('y')
plt.title('accuracy graph')

plt.plot(range(len(history.history['val_acc'])), history.history['val_acc'])

plt.show()

7.Predict

y_pred = model.predict_classes(test) ## predict, = predict

np.savetxt('CNN_6.csv', np.c_[range(1,len(test)+1),y_pred], delimiter=',', 
           header = 'ImageId,Label', comments = '', fmt='%d')

Well, Why not upload your CNN_6.csv to kaggle to see how the result goes . If the params are properly given , the result should be good enough.