DigitRecongnizer_CNN_DeepLearning

import numpy as np
import pandas as pd
%matplotlib inline
import matplotlib.pyplot as plt
import matplotlib.cm as cm

from lasagne.layers import  Conv2DLayer
from lasagne.layers import  MaxPool2DLayer
from lasagne.layers import  InputLayer
from lasagne.layers import  DenseLayer
from lasagne.layers import  DropoutLayer
from lasagne.nonlinearities import softmax
from lasagne.layers import  get_all_params
from lasagne.updates import adam

from nolearn.lasagne import NeuralNet
from nolearn.lasagne import TrainSplit
from nolearn.lasagne import objective
from nolearn.lasagne import PrintLayerInfo

data_set=pd.read_csv('./input/train.csv')
test_set=pd.read_csv('./input/test.csv')

data_set.info()

data_set.head()

	label	pixel0	pixel1	pixel2	pixel3	pixel4	pixel5	pixel6	pixel7	pixel8	…	pixel774	pixel775	pixel776	pixel777	pixel778	pixel779	pixel780	pixel781	pixel782	pixel783
0	1	0	0	0	0	0	0	0	0	0	…	0	0	0	0	0	0	0	0	0	0
1	0	0	0	0	0	0	0	0	0	0	…	0	0	0	0	0	0	0	0	0	0
2	1	0	0	0	0	0	0	0	0	0	…	0	0	0	0	0	0	0	0	0	0
3	4	0	0	0	0	0	0	0	0	0	…	0	0	0	0	0	0	0	0	0	0
4	0	0	0	0	0	0	0	0	0	0	…	0	0	0	0	0	0	0	0	0	0

5 rows × 785 columns

train_y=data_set[[0]]#get the training target
train_y.head()
train_target=train_y.values

train_target=train_target.ravel()
train_target.shape

(42000,)

train_input=data_set.iloc[:,1:].values#get the training input
train_input.shape

(42000, 784)

figs,axes=plt.subplots(4,4,figsize=(6,6))#show the first sixteen images and  their labels
for i in range(4):
    for j in range(4):
        axes[i,j].imshow(-train_input[i*4+j].reshape(28,28),cmap='gray')
        axes[i,j].set_xticks([])
        axes[i,j].set_yticks([])
        axes[i,j].set_title("label:{}".format(train_target[i*4+j]))

#change the shape of train_input
train_input=train_input.reshape([-1,1,28,28])
train_input.shape

(42000, 1, 28, 28)

#set the layers of neuralnetwork 
layers0=[
    (InputLayer,{'shape':(None,train_input.shape[1],train_input.shape[2],train_input.shape[3])}),
    (Conv2DLayer,{'num_filters':32,'filter_size':[3,3]}),
    (Conv2DLayer,{'num_filters':32,'filter_size':[3,3]}),
    (MaxPool2DLayer,{'pool_size':[2,2]}),
    (DenseLayer,{'num_units':100}),
    (DropoutLayer,{'p':0.2}),
    (DenseLayer,{'num_units':10,'nonlinearity':softmax})


]
net0=NeuralNet(layers=layers0,
               update_learning_rate=0.002,
               max_epochs=10,
               update=adam,
               objective_l2=0.0025,
               train_split=TrainSplit(eval_size=0.01)
              )

train_target=np.array(train_target,np.int64)

train_target.shape

(42000,)

train_target.dtype
train_target=train_target.astype(np.int32)

train_target.dtype

dtype('int32')

net0.fit(train_input,train_target)

test_set.shape

(28000, 784)

test_x=test_set.values

test_x=test_x.reshape(-1,1,28,28)
test_x.shape

(28000, 1, 28, 28)

test_y=net0.predict(test_x)

test_y=test_y.astype(np.int32)

test_y.dtype

dtype('int32')

figs1,axes1=plt.subplots(4,4,figsize=(6,6))
for i in range(4):
    for j in range(4):
        axes1[i,j].imshow(-test_x[i*4+j].reshape(28,28),cmap='gray')
        axes1[i,j].set_xticks([])
        axes1[i,j].set_yticks([])
        axes1[i,j].set_title('predict: {}'.format(test_y[i*4+j]))

len(test_y)

28000

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

from nolearn.lasagne.visualize import draw_to_notebook
from nolearn.lasagne.visualize import plot_loss
from nolearn.lasagne.visualize import plot_conv_weights
from nolearn.lasagne.visualize import plot_conv_activity
from nolearn.lasagne.visualize import plot_occlusion
from nolearn.lasagne.visualize import plot_saliency

draw_to_notebook(net0)

plot_loss(net0)

<module 'matplotlib.pyplot' from '/usr/lib/python3/dist-packages/matplotlib/pyplot.py'>

plot_occlusion(net0,train_input[:10],train_target[:10])

<module 'matplotlib.pyplot' from '/usr/lib/python3/dist-packages/matplotlib/pyplot.py'>

plot_saliency(net0,train_input[:5])

<module 'matplotlib.pyplot' from '/usr/lib/python3/dist-packages/matplotlib/pyplot.py'>

train