kaggle手写数字识别99.335%top12%

用卷积神经网络加上数据增强很容易达到99%以上准确率

调用包和导入数据

调用包

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import seaborn as sns
from keras.layers import BatchNormalization
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix
import itertools
from tensorflow.keras.callbacks import ReduceLROnPlateau,EarlyStopping
from keras import layers
from keras.utils.np_utils import to_categorical 
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D
from tensorflow.keras.optimizers import RMSprop
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import ReduceLROnPlateau

导入数据

sns.set(style='white', context='notebook', palette='deep')
train = pd.read_csv("换成你自己的路径")
test = pd.read_csv("换成你自己的路径")

数据预处理和划分训练集和验证集

数据预处理

除以255作归一化处理，调整训练集，变成4D张量，reshape中参数为-1的那个维度是自动调整，因为是灰白图片所以最后一个维度为1

Y_train = train["label"]
X_train = train.drop(labels=["label"],axis = 1)
del train
X_train = X_train / 255.0
test = test / 255.0
X_train = X_train.values.reshape(-1,28,28,1)
test = test.values.reshape(-1,28,28,1)
Y_train = to_categorical(Y_train, num_classes=10)

划分训练集和验证集

X_train, X_val, Y_train, Y_val = train_test_split(X_train, Y_train, test_size = 0.1,stratify = Y_train)

stratify参数设为True可以保证在训练集和验证集中数据的分布大致相同，在不平衡数据中非常实用

建模

构建模型

model = Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.5))
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.5))
model.add(Flatten())
model.add(Dense(256, activation="relu"))
model.add(Dense(10, activation = "softmax"))
optimizer = RMSprop(learning_rate=0.001, rho=0.9, epsilon=1e-08, decay=0.0)
model.compile(optimizer=optimizer, loss = "categorical_crossentropy", metrics=["accuracy"])

学习率衰减

learning_rate_reduction = ReduceLROnPlateau(monitor='val_acc',patience=3,verbose=1,factor=0.5,min_lr=0.00001)

数据增强

datagen = ImageDataGenerator(zca_whitening=True,rotation_range=8,  zoom_range = 0.2, 
                             width_shift_range=0.2,  height_shift_range=0.2,shear_range=0.2,
                             horizontal_flip=False, vertical_flip=False)  

这里进行了zca白化，用于减少数据冗余，使得特征之间相关性更低，horizontal_flip和vertical_flip都设置为False，5水平翻转可能会被当成2，6和9垂直翻转也会被误判，同时rotation_range不宜过高

训练模型

batch_size=256
epochs=150
early_stoping = EarlyStopping(monitor='val_loss',patience= 10,restore_best_weights=True,verbose=0)
history = model.fit_generator(datagen.flow(X_train,Y_train, batch_size=batch_size),
                              epochs = epochs, validation_data = (X_val,Y_val),
                              verbose = 2, steps_per_epoch=X_train.shape[0] // batch_size
                              , callbacks=[learning_rate_reduction,early_stoping])

这里设置了early_stoping，监控val_loss，10个轮次内没用继续降低就停止，同时储存最佳权重，X_train。shape[0]//batch_size防止生成的数据不足以跑完整个流程

提交文件

results = model.predict(test)
results = np.argmax(results,axis = 1)
results = pd.Series(results,name="Label")
submission = pd.concat([pd.Series(range(1,28001),name = "ImageId"),results],axis = 1)
submission.to_csv("cnn_mnist_datagen.csv",index=False)