PyTorch构建神经网络时y_train转张量类型错误的解决方法

问题根源 #

1. 威斯康星乳腺癌数据集的目标变量（diagnosis）是字符串类型（M=恶性，B=良性），并非你所说的数值型，直接转float32会触发类型错误。
2. 单独对y_train执行dropna()，会导致X_train和y_train样本数量不一致，后续批量训练会出现索引不匹配问题。
3. 未编码的字符串目标变量会让y_train.values成为numpy.object_类型，PyTorch不支持直接转换该类型为张量。

修复步骤 #

1. 对目标变量做数值编码 #

将字符串诊断结果转换为0/1数值：

# 替换原y定义部分
y = pd.DataFrame(breast_cancer_wisconsin_diagnostic.data.targets)
# 良性B转0，恶性M转1
y['diagnosis'] = y['diagnosis'].map({'B': 0, 'M': 1})

2. 同步处理缺失值 #

不要单独删除y的缺失样本，拆分数据集前统一处理所有缺失值：

# 合并特征与目标变量后统一删缺失值
df = pd.concat([X, y], axis=1)
df = df.dropna()
X = df.drop('diagnosis', axis=1)
y = df['diagnosis']

# 再执行数据集拆分
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

3. 正确转换为PyTorch张量 #

此时y_train已为数值型，用to_numpy()转换更稳妥：

y_train_tensor = torch.tensor(y_train.to_numpy(), dtype=torch.float32).view(-1, 1)
y_test_tensor = torch.tensor(y_test.to_numpy(), dtype=torch.float32).view(-1, 1)

完整修改后的代码 #

import pandas as pd
import torch
import torch.nn as nn
import torch.optim as optim
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from ucimlrepo import fetch_ucirepo  

breast_cancer_wisconsin_diagnostic = fetch_ucirepo(id=17) 
X = pd.DataFrame(breast_cancer_wisconsin_diagnostic.data.features)
y = pd.DataFrame(breast_cancer_wisconsin_diagnostic.data.targets)

# 关键修改1：对目标变量做数值编码
y['diagnosis'] = y['diagnosis'].map({'B': 0, 'M': 1})

# 关键修改2：统一处理缺失值
df = pd.concat([X, y], axis=1)
df = df.dropna()
X = df.drop('diagnosis', axis=1)
y = df['diagnosis']

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

X_train_tensor = torch.tensor(X_train, dtype=torch.float32)
X_test_tensor = torch.tensor(X_test, dtype=torch.float32) 

# 关键修改3：正确转换张量
y_train_tensor = torch.tensor(y_train.to_numpy(), dtype=torch.float32).view(-1, 1)
y_test_tensor = torch.tensor(y_test.to_numpy(), dtype=torch.float32).view(-1, 1)

class NeuralNetwork(nn.Module):
    def __init__(self):
        super(NeuralNetwork, self).__init__()
        self.fc1 = nn.Linear(X_train.shape[1], 16)
        self.fc2 = nn.Linear(16, 8)
        self.fc3 = nn.Linear(8, 1)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = torch.sigmoid(self.fc3(x))
        return x

model = NeuralNetwork()
loss_fn = nn.BCELoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)

n_epochs = 100
batch_size = 10

for epoch in range(n_epochs):
    model.train()
    for i in range(0, len(X_train_tensor), batch_size):
        X_batch = X_train_tensor[i:i+batch_size]
        y_batch = y_train_tensor[i:i+batch_size]

        optimizer.zero_grad()
        y_pred = model(X_batch)
        loss = loss_fn(y_pred, y_batch)
        loss.backward()
        optimizer.step()

    print(f'Finished epoch {epoch+1}, latest loss: {loss.item():.4f}')

model.eval()
with torch.no_grad():
    y_test_pred = model(X_test_tensor)
    y_test_pred_classes = (y_test_pred > 0.5).float()

accuracy = (y_test_pred_classes.eq(y_test_tensor).sum().item()) / len(y_test_tensor)
print(f'Accuracy on test set: {accuracy * 100:.2f}%') 

内容的提问来源于stack exchange，提问作者bahar homayoun