python - Pytorch 预测变得相同
问题描述
我开发了一个分类深度学习模型，我在训练时得到了这个结果：
tensor([[ 0.0141, -0.0279, 0.0623],
[ 0.0141, -0.0279, 0.0623],
[ 0.0141, -0.0279, 0.0623],
[ 0.0141, -0.0279, 0.0623]], grad_fn=<AddmmBackward>) tensor([1., 0., 1., 2.])
epoch = 1/10, step = 100/595, loss = 1.1103
tensor([[ 0.0294, -0.0161, 0.0353],
[ 0.0294, -0.0161, 0.0353],
[ 0.0294, -0.0161, 0.0353],
[ 0.0294, -0.0161, 0.0353]], grad_fn=<AddmmBackward>) tensor([0., 0., 0., 1.])
epoch = 1/10, step = 200/595, loss = 1.0971
tensor([[ 0.0315, -0.0199, 0.0369],
[ 0.0315, -0.0199, 0.0369],
[ 0.0315, -0.0199, 0.0369],
[ 0.0315, -0.0199, 0.0369]], grad_fn=<AddmmBackward>) tensor([0., 0., 2., 1.])
epoch = 1/10, step = 300/595, loss = 1.0951
tensor([[ 0.0388, -0.0036, 0.0134],
[ 0.0388, -0.0036, 0.0134],
[ 0.0388, -0.0036, 0.0134],
[ 0.0388, -0.0036, 0.0134]], grad_fn=<AddmmBackward>) tensor([1., 1., 1., 1.])
epoch = 1/10, step = 400/595, loss = 1.1186
tensor([[ 0.0476, -0.0057, 0.0066],
[ 0.0476, -0.0057, 0.0066],
[ 0.0476, -0.0057, 0.0066],
[ 0.0476, -0.0057, 0.0066]], grad_fn=<AddmmBackward>) tensor([1., 1., 2., 0.])
epoch = 1/10, step = 500/595, loss = 1.1043
tensor([[-0.0150, 0.0246, 0.0390],
[-0.0150, 0.0246, 0.0390],
[-0.0150, 0.0246, 0.0390],
[-0.0150, 0.0246, 0.0390]], grad_fn=<AddmmBackward>) tensor([2., 1., 2., 0.])
epoch = 2/10, step = 100/595, loss = 1.0932
tensor([[0.0168, 0.0212, 0.0106],
[0.0168, 0.0212, 0.0106],
[0.0168, 0.0212, 0.0106],
[0.0168, 0.0212, 0.0106]], grad_fn=<AddmmBackward>) tensor([0., 1., 0., 0.])
epoch = 2/10, step = 200/595, loss = 1.0969
tensor([[ 0.0448, -0.0110, 0.0148],
[ 0.0448, -0.0110, 0.0148],
[ 0.0448, -0.0110, 0.0148],
[ 0.0448, -0.0110, 0.0148]], grad_fn=<AddmmBackward>) tensor([0., 2., 2., 1.])
epoch = 2/10, step = 300/595, loss = 1.0992
tensor([[0.0310, 0.0007, 0.0168],
[0.0310, 0.0007, 0.0168],
[0.0310, 0.0007, 0.0168],
[0.0310, 0.0007, 0.0168]], grad_fn=<AddmmBackward>) tensor([1., 1., 0., 0.])
epoch = 2/10, step = 400/595, loss = 1.0990
tensor([[ 0.0336, 0.0241, -0.0092],
[ 0.0336, 0.0241, -0.0092],
[ 0.0336, 0.0241, -0.0092],
[ 0.0336, 0.0241, -0.0092]], grad_fn=<AddmmBackward>) tensor([0., 1., 1., 1.])
epoch = 2/10, step = 500/595, loss = 1.0885
tensor([[0.0288, 0.0057, 0.0140],
[0.0288, 0.0057, 0.0140],
[0.0288, 0.0057, 0.0140],
[0.0288, 0.0057, 0.0140]], grad_fn=<AddmmBackward>) tensor([1., 0., 1., 1.])
epoch = 3/10, step = 100/595, loss = 1.1033
tensor([[0.0243, 0.0002, 0.0240],
[0.0243, 0.0002, 0.0240],
[0.0243, 0.0002, 0.0240],
[0.0243, 0.0002, 0.0240]], grad_fn=<AddmmBackward>) tensor([2., 2., 1., 1.])
epoch = 3/10, step = 200/595, loss = 1.1027
tensor([[ 0.0512, 0.0081, -0.0107],
[ 0.0512, 0.0081, -0.0107],
[ 0.0512, 0.0081, -0.0107],
[ 0.0512, 0.0081, -0.0107]], grad_fn=<AddmmBackward>) tensor([2., 0., 0., 1.])
epoch = 3/10, step = 300/595, loss = 1.0902
tensor([[ 0.0374, -0.0038, 0.0150],
[ 0.0374, -0.0038, 0.0150],
[ 0.0374, -0.0038, 0.0150],
[ 0.0374, -0.0038, 0.0150]], grad_fn=<AddmmBackward>) tensor([0., 2., 0., 0.])
epoch = 3/10, step = 400/595, loss = 1.0831
tensor([[ 0.0325, 0.0225, -0.0065],
[ 0.0325, 0.0225, -0.0065],
[ 0.0325, 0.0225, -0.0065],
[ 0.0325, 0.0225, -0.0065]], grad_fn=<AddmmBackward>) tensor([1., 2., 2., 1.])
epoch = 3/10, step = 500/595, loss = 1.1069
tensor([[0.0210, 0.0100, 0.0175],
[0.0210, 0.0100, 0.0175],
[0.0210, 0.0100, 0.0175],
[0.0210, 0.0100, 0.0175]], grad_fn=<AddmmBackward>) tensor([0., 1., 1., 2.])
epoch = 4/10, step = 100/595, loss = 1.1002
tensor([[0.0189, 0.0051, 0.0245],
[0.0189, 0.0051, 0.0245],
[0.0189, 0.0051, 0.0245],
[0.0189, 0.0051, 0.0245]], grad_fn=<AddmmBackward>) tensor([0., 2., 0., 2.])
epoch = 4/10, step = 200/595, loss = 1.0931
tensor([[0.0184, 0.0076, 0.0226],
[0.0184, 0.0076, 0.0226],
[0.0184, 0.0076, 0.0226],
[0.0184, 0.0076, 0.0226]], grad_fn=<AddmmBackward>) tensor([1., 2., 1., 0.])
epoch = 4/10, step = 300/595, loss = 1.1008
tensor([[-0.0005, 0.0347, 0.0144],
[-0.0005, 0.0347, 0.0144],
[-0.0005, 0.0347, 0.0144],
[-0.0005, 0.0347, 0.0144]], grad_fn=<AddmmBackward>) tensor([1., 1., 0., 1.])
epoch = 4/10, step = 400/595, loss = 1.0890
tensor([[0.0122, 0.0191, 0.0173],
[0.0122, 0.0191, 0.0173],
[0.0122, 0.0191, 0.0173],
[0.0122, 0.0191, 0.0173]], grad_fn=<AddmmBackward>) tensor([2., 1., 2., 2.])
epoch = 4/10, step = 500/595, loss = 1.0971
tensor([[ 0.0323, -0.0013, 0.0175],
[ 0.0323, -0.0013, 0.0175],
[ 0.0323, -0.0013, 0.0175],
[ 0.0323, -0.0013, 0.0175]], grad_fn=<AddmmBackward>) tensor([2., 0., 0., 2.])
epoch = 5/10, step = 100/595, loss = 1.0900
tensor([[ 0.0335, 0.0183, -0.0032],
[ 0.0335, 0.0183, -0.0032],
[ 0.0335, 0.0183, -0.0032],
[ 0.0335, 0.0183, -0.0032]], grad_fn=<AddmmBackward>) tensor([1., 2., 1., 2.])
epoch = 5/10, step = 200/595, loss = 1.1074
tensor([[0.0239, 0.0134, 0.0112],
[0.0239, 0.0134, 0.0112],
[0.0239, 0.0134, 0.0112],
[0.0239, 0.0134, 0.0112]], grad_fn=<AddmmBackward>) tensor([1., 2., 0., 2.])
epoch = 5/10, step = 300/595, loss = 1.0999
tensor([[ 0.0394, 0.0244, -0.0153],
[ 0.0394, 0.0244, -0.0153],
[ 0.0394, 0.0244, -0.0153],
[ 0.0394, 0.0244, -0.0153]], grad_fn=<AddmmBackward>) tensor([1., 1., 2., 2.])
epoch = 5/10, step = 400/595, loss = 1.1105
tensor([[ 0.0529, 0.0281, -0.0324],
[ 0.0529, 0.0281, -0.0324],
[ 0.0529, 0.0281, -0.0324],
[ 0.0529, 0.0281, -0.0324]], grad_fn=<AddmmBackward>) tensor([2., 2., 2., 2.])
epoch = 5/10, step = 500/595, loss = 1.1478
tensor([[0.0104, 0.0371, 0.0011],
[0.0104, 0.0371, 0.0011],
[0.0104, 0.0371, 0.0011],
[0.0104, 0.0371, 0.0011]], grad_fn=<AddmmBackward>) tensor([2., 1., 1., 1.])
epoch = 6/10, step = 100/595, loss = 1.0868
tensor([[0.0038, 0.0232, 0.0215],
[0.0038, 0.0232, 0.0215],
[0.0038, 0.0232, 0.0215],
[0.0038, 0.0232, 0.0215]], grad_fn=<AddmmBackward>) tensor([2., 0., 1., 2.])
epoch = 6/10, step = 200/595, loss = 1.0973
tensor([[0.0094, 0.0273, 0.0119],
[0.0094, 0.0273, 0.0119],
[0.0094, 0.0273, 0.0119],
[0.0094, 0.0273, 0.0119]], grad_fn=<AddmmBackward>) tensor([1., 0., 0., 2.])
epoch = 6/10, step = 300/595, loss = 1.1003
tensor([[ 0.0397, -0.0106, 0.0194],
[ 0.0397, -0.0106, 0.0194],
[ 0.0397, -0.0106, 0.0194],
[ 0.0397, -0.0106, 0.0194]], grad_fn=<AddmmBackward>) tensor([0., 2., 2., 1.])
epoch = 6/10, step = 400/595, loss = 1.0980
tensor([[ 0.0321, -0.0099, 0.0264],
[ 0.0321, -0.0099, 0.0264],
[ 0.0321, -0.0099, 0.0264],
[ 0.0321, -0.0099, 0.0264]], grad_fn=<AddmmBackward>) tensor([1., 1., 2., 2.])
epoch = 6/10, step = 500/595, loss = 1.1067
tensor([[ 0.0487, -0.0066, 0.0064],
[ 0.0487, -0.0066, 0.0064],
[ 0.0487, -0.0066, 0.0064],
[ 0.0487, -0.0066, 0.0064]], grad_fn=<AddmmBackward>) tensor([2., 0., 0., 1.])
epoch = 7/10, step = 100/595, loss = 1.0908
tensor([[ 0.0628, 0.0229, -0.0372],
[ 0.0628, 0.0229, -0.0372],
[ 0.0628, 0.0229, -0.0372],
[ 0.0628, 0.0229, -0.0372]], grad_fn=<AddmmBackward>) tensor([2., 2., 0., 0.])
epoch = 7/10, step = 200/595, loss = 1.1028
tensor([[ 0.0379, 0.0228, -0.0121],
[ 0.0379, 0.0228, -0.0121],
[ 0.0379, 0.0228, -0.0121],
[ 0.0379, 0.0228, -0.0121]], grad_fn=<AddmmBackward>) tensor([0., 0., 0., 0.])
epoch = 7/10, step = 300/595, loss = 1.0771
tensor([[ 0.0359, 0.0242, -0.0116],
[ 0.0359, 0.0242, -0.0116],
[ 0.0359, 0.0242, -0.0116],
[ 0.0359, 0.0242, -0.0116]], grad_fn=<AddmmBackward>) tensor([1., 1., 0., 1.])
epoch = 7/10, step = 400/595, loss = 1.0879
tensor([[ 0.0295, 0.0306, -0.0115],
[ 0.0295, 0.0306, -0.0115],
[ 0.0295, 0.0306, -0.0115],
[ 0.0295, 0.0306, -0.0115]], grad_fn=<AddmmBackward>) tensor([1., 0., 2., 1.])
epoch = 7/10, step = 500/595, loss = 1.0952
tensor([[0.0272, 0.0016, 0.0197],
[0.0272, 0.0016, 0.0197],
[0.0272, 0.0016, 0.0197],
[0.0272, 0.0016, 0.0197]], grad_fn=<AddmmBackward>) tensor([2., 1., 2., 0.])
epoch = 8/10, step = 100/595, loss = 1.0978
tensor([[0.0273, 0.0187, 0.0025],
[0.0273, 0.0187, 0.0025],
[0.0273, 0.0187, 0.0025],
[0.0273, 0.0187, 0.0025]], grad_fn=<AddmmBackward>) tensor([1., 2., 1., 0.])
epoch = 8/10, step = 200/595, loss = 1.0980
tensor([[ 0.0573, 0.0246, -0.0333],
[ 0.0573, 0.0246, -0.0333],
[ 0.0573, 0.0246, -0.0333],
[ 0.0573, 0.0246, -0.0333]], grad_fn=<AddmmBackward>) tensor([1., 0., 0., 0.])
epoch = 8/10, step = 300/595, loss = 1.0664
tensor([[ 0.0169, 0.0369, -0.0052],
[ 0.0169, 0.0369, -0.0052],
[ 0.0169, 0.0369, -0.0052],
[ 0.0169, 0.0369, -0.0052]], grad_fn=<AddmmBackward>) tensor([2., 2., 1., 2.])
epoch = 8/10, step = 400/595, loss = 1.1096
tensor([[0.0025, 0.0081, 0.0379],
[0.0025, 0.0081, 0.0379],
[0.0025, 0.0081, 0.0379],
[0.0025, 0.0081, 0.0379]], grad_fn=<AddmmBackward>) tensor([0., 1., 1., 0.])
epoch = 8/10, step = 500/595, loss = 1.1096
tensor([[-0.0015, -0.0033, 0.0534],
[-0.0015, -0.0033, 0.0534],
[-0.0015, -0.0033, 0.0534],
[-0.0015, -0.0033, 0.0534]], grad_fn=<AddmmBackward>) tensor([1., 0., 0., 1.])
epoch = 9/10, step = 100/595, loss = 1.1176
tensor([[0.0130, 0.0270, 0.0085],
[0.0130, 0.0270, 0.0085],
[0.0130, 0.0270, 0.0085],
[0.0130, 0.0270, 0.0085]], grad_fn=<AddmmBackward>) tensor([0., 0., 0., 1.])
epoch = 9/10, step = 200/595, loss = 1.0983
tensor([[ 0.0625, 0.0357, -0.0496],
[ 0.0625, 0.0357, -0.0496],
[ 0.0625, 0.0357, -0.0496],
[ 0.0625, 0.0357, -0.0496]], grad_fn=<AddmmBackward>) tensor([1., 2., 0., 1.])
epoch = 9/10, step = 300/595, loss = 1.0949
tensor([[ 0.0622, -0.0005, -0.0132],
[ 0.0622, -0.0005, -0.0132],
[ 0.0622, -0.0005, -0.0132],
[ 0.0622, -0.0005, -0.0132]], grad_fn=<AddmmBackward>) tensor([1., 0., 0., 2.])
epoch = 9/10, step = 400/595, loss = 1.0876
tensor([[ 0.0217, -0.0025, 0.0293],
[ 0.0217, -0.0025, 0.0293],
[ 0.0217, -0.0025, 0.0293],
[ 0.0217, -0.0025, 0.0293]], grad_fn=<AddmmBackward>) tensor([1., 0., 2., 0.])
epoch = 9/10, step = 500/595, loss = 1.0973
tensor([[ 0.0379, -0.0200, 0.0306],
[ 0.0379, -0.0200, 0.0306],
[ 0.0379, -0.0200, 0.0306],
[ 0.0379, -0.0200, 0.0306]], grad_fn=<AddmmBackward>) tensor([1., 0., 2., 1.])
epoch = 10/10, step = 100/595, loss = 1.1080
tensor([[ 0.0466, 0.0045, -0.0025],
[ 0.0466, 0.0045, -0.0025],
[ 0.0466, 0.0045, -0.0025],
[ 0.0466, 0.0045, -0.0025]], grad_fn=<AddmmBackward>) tensor([2., 2., 1., 1.])
epoch = 10/10, step = 200/595, loss = 1.1141
tensor([[0.0321, 0.0124, 0.0040],
[0.0321, 0.0124, 0.0040],
[0.0321, 0.0124, 0.0040],
[0.0321, 0.0124, 0.0040]], grad_fn=<AddmmBackward>) tensor([2., 0., 1., 2.])
epoch = 10/10, step = 300/595, loss = 1.1017
tensor([[0.0204, 0.0120, 0.0162],
[0.0204, 0.0120, 0.0162],
[0.0204, 0.0120, 0.0162],
[0.0204, 0.0120, 0.0162]], grad_fn=<AddmmBackward>) tensor([1., 2., 0., 0.])
epoch = 10/10, step = 400/595, loss = 1.0976
tensor([[0.0092, 0.0228, 0.0165],
[0.0092, 0.0228, 0.0165],
[0.0092, 0.0228, 0.0165],
[0.0092, 0.0228, 0.0165]], grad_fn=<AddmmBackward>) tensor([1., 2., 0., 0.])
epoch = 10/10, step = 500/595, loss = 1.1004
acc = 34.22818791946309
Accuracy of 0: 0.0, correct predictions = 0
Accuracy of 1: 0.0, correct predictions = 0
Accuracy of 2: 100.0, correct predictions = 204
Process finished with exit code 0
无论输入是什么，它都会给我相同的预测
这是模型：
import torch
import torchvision
from sklearn.model_selection import train_test_split
from torch.utils.data import Dataset
from torchvision.transforms import transforms
import torch.nn as nn
import pandas as pd
import numpy as np
num_classes = 3
batch_size = 4
hidden_size = 100
input_size = 196
learning_rate = 0.01
num_epochs = 10
class price_dataset():
def __init__(self):
xy = pd.read_csv('data_balanced.csv')
self.n_samples = xy.shape[0]
xy = xy.to_numpy()
self.x_data = torch.from_numpy(xy[:, 7:203].astype(np.float32))
self.y_data = torch.from_numpy(xy[:, 6].astype(np.float32))
def __getitem__(self, index):
return self.x_data[index], self.y_data[index]
def __len__(self):
return self.n_samples
dataset = price_dataset()
train_dataset, test_dataset = train_test_split(dataset, test_size=0.20, random_state=0)
# Data loader
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=batch_size,
shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=batch_size,
shuffle=False)
# print(example_data)
class NeuralNet(nn.Module):
def __init__(self, input_size, hidden_size, num_classes):
super(NeuralNet, self).__init__()
self.input_size = input_size
self.l1 = nn.Linear(input_size, hidden_size)
self.relu = nn.ReLU()
self.l2 = nn.Linear(hidden_size, num_classes)
def forward(self, inputs):
out = self.l1(inputs)
out = self.relu(out)
out = self.l2(out)
return out
model = NeuralNet(input_size, hidden_size, num_classes)
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)
n_total_steps = len(train_loader)
for epoch in range(num_epochs):
for i, (input, labels) in enumerate(train_loader):
y_pred = model(input)
loss = criterion(y_pred, labels.long())
optimizer.zero_grad()
loss.backward()
optimizer.step()
# print(f'prediction = {y_pred[0]}, label = {labels[0]}')
if (i + 1) % 100 == 0:
print(y_pred, labels)
print(f'epoch = {epoch + 1}/{num_epochs}, step = {i + 1}/{n_total_steps}, loss = {loss.item():.4f}')
with torch.no_grad():
n_correct = 0
n_samples = 0
n_class_correct = [0 for i in range(3)]
n_class_samples = [0 for i in range(3)]
for input, labels in test_loader:
labels = torch.flatten(labels)
y_pred = model(input)
_, prediction = torch.max(y_pred.data, 1)
# print(f'prediction = {prediction[0]}, label = {labels[0]}')
n_samples += labels.shape[0]
# n_correct += (prediction == labels).sum().item()
for i in range(batch_size):
label = labels[i]
pred = prediction[i]
if (label==pred):
n_class_correct[int(label)] +=1
n_class_samples[int(label)] += 1
for pred, label in zip(prediction, labels):
# print(f'prediction = {pred}, label = {label}')
if pred != 0 and label !=0 and pred == label:
n_correct += (pred == label).item()
acc = 100 * (n_correct / n_samples)
print(f'acc = {acc}')
for i in range(3):
acc = 100 * n_class_correct[i]/n_class_samples[i]
print(f'Accuracy of {i}: {acc}, correct predictions = {n_class_correct[i]}')
模型有问题吗？我尝试更改优化器和学习率，但没有任何改变我也尝试更改隐藏层大小，但它也没有解决问题
解决方案
推荐阅读
- arrays - 我可以使用矢量化操作在没有循环的情况下重复 random.choice
- javascript - 如何将数据发送到另一个 NextJS 页面？
- powerbi - 如何在 PowerBI 中计算员工人数
- reactjs - ReactDnd-无法将状态从 useDrag 发送到 useDrop
- flutter - 是否有错误的解决方案：参数类型'Stream
- flutter - 如何在 Flutter 中将小部件设置为按钮
- python - 用其他值替换空值
- python - 使用“自定义用户模型”（Django）条纹“webhook”错误
- javascript - 本机代码方法返回未定义而不是字符串
- dataframe - 计算其值落入 SPARK 中每个 bin 的记录数