玩了下kaggle,采用了Dense和dropout,增大batch_size和epochs,提交结果准确率为 0.99800(5%),哈哈。
# -*- coding: utf-8 -*-
import tensorflow as tf
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
#from keras.optimizers import SGD
# -*- coding: utf-8 -*-
import numpy as np
import csv
def LoadTraincsvfile():
tmp=np.loadtxt('./data/train.csv',dtype=np.str,delimiter=',')
x_data=tmp[1:,1:].astype(np.float)
y_data=tmp[1:,0].astype(np.float)
return x_data,y_data
x_data,y_data=LoadTraincsvfile()
path = './data/mnist.npz'
f = np.load(path)
x_train1, y_train1 = f['x_train'], f['y_train']
x_test1, y_test1 = f['x_test'], f['y_test']
f.close()
print (x_data[0:41000,:].shape)
print (x_train1.shape)
x_train1 = x_train1.reshape(60000, 784).astype('float32')
x_train=np.concatenate((x_data[0:42000,:],x_train1),axis=0)
y_train=np.concatenate((y_data[0:42000],y_train1),axis=0)
x_test=x_data[41000:,:]
y_test=y_data[41000:]
print (x_train.shape)
print (y_train.shape)
x_train /= 255
x_test /= 255
print(x_train.shape, 'train samples')
print(x_test.shape, 'test samples')
y_train = keras.utils.to_categorical(y_train, num_classes=10)
y_test = keras.utils.to_categorical(y_test, num_classes=10)
model = Sequential()
model.add(Dense(200, activation='relu', input_dim=784))
model.add(Dense(120, activation='relu'))
model.add(Dense(120, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))
rms = keras.optimizers.Adam(lr=0.0005, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
model.compile(loss='binary_crossentropy',
optimizer=rms,
metrics=['accuracy'])
model.fit(x_train, y_train,
epochs=300,
batch_size=512)
score = model.evaluate(x_test, y_test, batch_size=4096)
print(model.metrics_names)
print(score)
#预测部分
is_predict=1
if(is_predict):
tmp = np.loadtxt('./data/test.csv', dtype=np.str, delimiter=',')
x_predict = tmp[1:, :].astype(np.float)
x_predict /= 255
classes = model.predict_classes(x_predict)
resultlist = []
imgid = 1
for i in classes:
res_data = {'ImageId': imgid, 'Label': i}
resultlist.append(res_data)
imgid += 1
headers = ['ImageId', 'Label']
with open('result.csv', 'w', newline='') as f:
# 标头在这里传入,作为第一行数据
writer = csv.DictWriter(f, headers)
writer.writeheader()
for row in resultlist:
writer.writerow(row)