题目地址:sofasofa-6
卷积神经网络(Pytorch, Python)
跟官方答案keras思路一样,简化了一下,毕竟我太菜了。答案里把训练集30%用作测试集,还画了图,我都省略了。
注意点
torch和numpy的格式不一样,numpy转torch需要torch数组=torch.from_numpy(numpy数组),转化完之后加.float(),因为发现神经网络的参数类型和训练数据一样的,本来是long,weight也会变成Long,然后报错说weight期望的数据类型是long结果竟然是float。文中对应
#将numpy格式转换为torch格式,并且要变成float,不然说weight期望是long,却是float的,报错
train_x,train_y=torch.from_numpy(train_x).float(),torch.from_numpy(train_y).float()
主要思路
建网络
def build_model():#输入数据 图片(1,40,40)
net=nn.Sequential(#除起来截断#数据形状(n,高,长,宽)
nn.Conv2d(in_channels=1,out_channels=8,kernel_size=5),#40-4=36
nn.ReLU(),
nn.Conv2d(in_channels=8,out_channels=16,kernel_size=3),#36-2=34
nn.ReLU(),
nn.MaxPool2d(kernel_size=4),#34/4=8
nn.Conv2d(16,16,3),#8-2=6
nn.ReLU(),
nn.MaxPool2d(4),#6/4=1
nn.Flatten(),#16
nn.Linear(16,128),
nn.Dropout(0.5),
nn.Linear(128,1),
nn.Sigmoid()
)
return net
dataset和dataloader处理
train_x和train_y之前都经过处理了,这里省略,都变成了numpy类型的矩阵。
#将numpy格式转换为torch格式,并且要变成float,不然说weight期望是long,却是float的,报错
train_x,train_y=torch.from_numpy(train_x).float(),torch.from_numpy(train_y).float()
#简易的创建一个torch格式的dataset,网上的都是写类创建的,当然这个函数的实现也是写类
torch_dataset=Data.TensorDataset(train_x,train_y)
data_loader=Data.DataLoader(
dataset=torch_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=2,
)
训练和预测
optimizer=torch.optim.Adam(model.parameters(),lr=0.001)
# training and testing
for epoch in range(epochs):
for step,(b_x,b_y) in enumerate(data_loader): # 分配 batch data, normalize x when iterate train_loader
output = model(b_x) # cnn output
loss=F.binary_cross_entropy(output,b_y) # cross entropy loss
optimizer.zero_grad() # clear gradients for this training step
loss.backward() # backpropagation, compute gradients
optimizer.step() # apply gradients
test=torch.from_numpy(test).float()
pred=model(test).detach().numpy()#报错后就这么提醒我的,加.detach().numpy()
完整代码
import pandas as pd
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data as Data
def load_train_test_data(train,test):
np.random.shuffle(train)
labels=train[:,-1]
data_test=np.array(test)
data, data_test = data_modify_suitable_train(train, True), data_modify_suitable_train(test, False)
return data, labels,data_test
def data_modify_suitable_train(data_set=None, type=True):
if data_set is not None:
data = []
if type is True:
np.random.shuffle(data_set)#总喜欢shuffle一下
data = data_set[:, 0: data_set.shape[1] - 1]
else:
data = data_set
data = np.array([np.reshape(i, (40, 40)) for i in data])#一维转化成二维
data = np.array([np.reshape(i, (1,i.shape[0], i.shape[1])) for i in data])#(高,长,宽)一般卷积都有个长宽高,加个高
return data
def f1(y_true, y_pred):
def recall(y_true,y_pred):
true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))
recall = true_positives / (possible_positives + K.epsilon())
return recall
def precision(y_true,y_pred):
true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))
precision = true_positives / (predicted_positives + K.epsilon())
return precision
precision = precision(y_true, y_pred)
recall = recall(y_true, y_pred)
return 2 * ((precision * recall) / (precision + recall))
def build_model():#输入数据 图片(1,40,40)
net=nn.Sequential(#除起来截断#数据形状(n,高,长,宽)
nn.Conv2d(in_channels=1,out_channels=8,kernel_size=5),#40-4=36
nn.ReLU(),
nn.Conv2d(in_channels=8,out_channels=16,kernel_size=3),#36-2=34
nn.ReLU(),
nn.MaxPool2d(kernel_size=4),#34/4=8
nn.Conv2d(16,16,3),#8-2=6
nn.ReLU(),
nn.MaxPool2d(4),#6/4=1
nn.Flatten(),#16
nn.Linear(16,128),
nn.Dropout(0.5),
nn.Linear(128,1),
nn.Sigmoid()
)
return net
def train_model(train,test,batch_size=64,epochs=10,model=None):
train_x, train_y, test = load_train_test_data(train, test)
if model is None:
model=build_model()
#将numpy格式转换为torch格式,并且要变成float,不然说weight期望是long,却是float的,报错
train_x,train_y=torch.from_numpy(train_x).float(),torch.from_numpy(train_y).float()
#简易的创建一个torch格式的dataset,网上的都是写类创建的,当然这个函数的实现也是写类
torch_dataset=Data.TensorDataset(train_x,train_y)
data_loader=Data.DataLoader(
dataset=torch_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=2,
)
optimizer=torch.optim.Adam(model.parameters(),lr=0.001)
# training and testing
for epoch in range(epochs):
for step,(b_x,b_y) in enumerate(data_loader): # 分配 batch data, normalize x when iterate train_loader
output = model(b_x) # cnn output
loss=F.binary_cross_entropy(output,b_y) # cross entropy loss
optimizer.zero_grad() # clear gradients for this training step
loss.backward() # backpropagation, compute gradients
optimizer.step() # apply gradients
test=torch.from_numpy(test).float()
pred=model(test).detach().numpy()#报错后就这么提醒我的,加.detach().numpy()
return pred
if __name__ == '__main__':
train, test = pd.read_csv('train.csv'), pd.read_csv('test.csv')
train = np.array(train.drop('id', axis=1))
test = np.array(test.drop('id', axis=1))
pred = train_model(train, test)
pred=(pred>0.5).astype(int)#转换格式
submit = pd.read_csv('sample_submit.csv')
submit['y'] = pred
submit.to_csv('Pytorch_my_CNN_prediction.csv', index=False)