refactor: unify model base and update docs roadmap
This commit is contained in:
parent
ecf6242fe7
commit
9f241757c6
|
|
@ -1,81 +1,96 @@
|
|||
import torch
|
||||
import torch.nn as nn
|
||||
from sklearn.preprocessing import LabelEncoder
|
||||
from torch.utils.data import DataLoader, TensorDataset
|
||||
import numpy as np
|
||||
|
||||
from Qtorch.Models.Qnn import Qnn
|
||||
|
||||
|
||||
class QCNN(Qnn):
|
||||
def __init__(self, X_train, y_train, X_test, y_test, labels=None, dropout_rate=0.3):
|
||||
super(QCNN, self).__init__()
|
||||
def __init__(
|
||||
self,
|
||||
data,
|
||||
labels=None,
|
||||
conv_channels=(16, 32),
|
||||
kernel_size=3,
|
||||
hidden_size=128,
|
||||
dropout_rate=0.3,
|
||||
test_size=0.2,
|
||||
random_state=None,
|
||||
batch_size=64,
|
||||
learning_rate=0.00001,
|
||||
weight_decay=1e-5,
|
||||
lr_scheduler_patience=10,
|
||||
early_stop_patience=100,
|
||||
early_stop_threshold=0.99,
|
||||
):
|
||||
super(QCNN, self).__init__(
|
||||
data=data,
|
||||
labels=labels,
|
||||
test_size=test_size,
|
||||
random_state=random_state,
|
||||
batch_size=batch_size,
|
||||
learning_rate=learning_rate,
|
||||
weight_decay=weight_decay,
|
||||
lr_scheduler_patience=lr_scheduler_patience,
|
||||
early_stop_patience=early_stop_patience,
|
||||
early_stop_threshold=early_stop_threshold,
|
||||
)
|
||||
|
||||
self.LABEL_ENCODER = LabelEncoder()
|
||||
self.conv_channels = tuple(conv_channels)
|
||||
self.kernel_size = kernel_size
|
||||
self.hidden_size = hidden_size
|
||||
self.dropout_rate = dropout_rate
|
||||
|
||||
self.X_train, self.y_train, self.X_test, self.y_test = X_train, y_train, X_test, y_test
|
||||
self.labels = labels
|
||||
self.feature_extractor = nn.Sequential()
|
||||
self.classifier = nn.Sequential()
|
||||
|
||||
input_size = X_train.shape[1] # 输入的长度
|
||||
num_classes = len(set(y_train)) # 分类数
|
||||
# 构造 1D CNN 网络结构
|
||||
self.build_model(input_shape=self.X_train.shape[1:], num_classes=self.num_classes)
|
||||
self._model_built = True
|
||||
|
||||
# 网络层：卷积层 + 池化层 + 全连接层
|
||||
self.layers = nn.ModuleList()
|
||||
self.layers.append(nn.Conv1d(in_channels=1, out_channels=16, kernel_size=3)) # 卷积层
|
||||
self.layers.append(nn.MaxPool1d(kernel_size=2)) # 池化层
|
||||
self.layers.append(nn.Conv1d(in_channels=16, out_channels=32, kernel_size=3)) # 卷积层
|
||||
self.layers.append(nn.MaxPool1d(kernel_size=2)) # 池化层
|
||||
def _transform_features(self, features):
|
||||
# 1D CNN 输入格式: [batch, channel=1, length]
|
||||
return torch.tensor(features, dtype=torch.float32).unsqueeze(1)
|
||||
|
||||
# 计算展平后的大小
|
||||
conv_output_size = self._get_conv_output_size(input_size) # 卷积后的输出大小
|
||||
print(f"Conv output size: {conv_output_size}") # 打印卷积后的输出大小
|
||||
self.layers.append(nn.Linear(conv_output_size, 128)) # 全连接层
|
||||
self.layers.append(nn.Linear(128, num_classes)) # 输出层
|
||||
def build_model(self, input_shape, num_classes):
|
||||
if len(self.conv_channels) == 0:
|
||||
raise ValueError("'conv_channels' must contain at least one channel size.")
|
||||
|
||||
input_length = int(np.prod(input_shape))
|
||||
|
||||
conv_layers = []
|
||||
in_channels = 1
|
||||
for out_channels in self.conv_channels:
|
||||
conv_layers.append(nn.Conv1d(in_channels=in_channels, out_channels=out_channels, kernel_size=self.kernel_size))
|
||||
conv_layers.append(nn.ReLU())
|
||||
conv_layers.append(nn.MaxPool1d(kernel_size=2))
|
||||
in_channels = out_channels
|
||||
self.feature_extractor = nn.Sequential(*conv_layers)
|
||||
|
||||
conv_output_size = self._get_conv_output_size(input_length)
|
||||
self.classifier = nn.Sequential(
|
||||
nn.Linear(conv_output_size, self.hidden_size),
|
||||
nn.ReLU(),
|
||||
nn.Dropout(self.dropout_rate),
|
||||
nn.Linear(self.hidden_size, num_classes),
|
||||
)
|
||||
|
||||
self.__init_weights()
|
||||
|
||||
def _get_conv_output_size(self, input_size):
|
||||
# 计算卷积后的输出尺寸
|
||||
x = torch.randn(1, 1, input_size) # 创建一个假的输入张量
|
||||
for layer in self.layers:
|
||||
x = layer(x) # 通过每一层
|
||||
return int(x.numel()) # 返回展平后的输出大小
|
||||
|
||||
def _get_conv_output_size(self, input_length):
|
||||
x = torch.randn(1, 1, input_length)
|
||||
x = self.feature_extractor(x)
|
||||
return int(x.numel())
|
||||
|
||||
def forward(self, x):
|
||||
# 通过卷积和池化层
|
||||
for layer in self.layers[:-2]: # 除去最后两个 Linear 层
|
||||
x = layer(x)
|
||||
|
||||
# 展平卷积后的输出
|
||||
x = x.view(x.size(0), -1) # 这样 x 会变成 (batch_size, conv_output_size)
|
||||
|
||||
# 通过全连接层
|
||||
x = self.layers[-2](x)
|
||||
x = self.layers[-1](x)
|
||||
|
||||
x = self.feature_extractor(x)
|
||||
x = x.view(x.size(0), -1)
|
||||
x = self.classifier(x)
|
||||
return x
|
||||
|
||||
def __init_weights(self):
|
||||
for m in self.modules():
|
||||
if isinstance(m, nn.Conv1d) or isinstance(m, nn.Linear):
|
||||
if isinstance(m, (nn.Conv1d, nn.Linear)):
|
||||
nn.init.xavier_uniform_(m.weight)
|
||||
if m.bias is not None:
|
||||
m.bias.data.fill_(0.01)
|
||||
|
||||
def _prepare_data(self):
|
||||
|
||||
# 将data转换为tensor形式, unsqueeze可以创建多一维度的，给
|
||||
X_train_tensor = torch.tensor(self.X_train, dtype=torch.float32).unsqueeze(1)
|
||||
self.y_train = self.LABEL_ENCODER.fit_transform(self.y_train)
|
||||
y_train_tensor = torch.tensor(self.y_train, dtype=torch.long)
|
||||
|
||||
X_test_tensor = torch.tensor(self.X_test, dtype=torch.float32).unsqueeze(1)
|
||||
self.y_test = self.LABEL_ENCODER.transform(self.y_test)
|
||||
y_test_tensor = torch.tensor(self.y_test, dtype=torch.long)
|
||||
|
||||
train_dataset = TensorDataset(X_train_tensor, y_train_tensor)
|
||||
test_dataset = TensorDataset(X_test_tensor, y_test_tensor)
|
||||
|
||||
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
|
||||
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)
|
||||
|
||||
return train_loader, test_loader
|
||||
nn.init.zeros_(m.bias)
|
||||
|
|
|
|||
|
|
@ -10,34 +10,63 @@ class Qmlp(Qnn):
|
|||
labels=None,
|
||||
dropout_rate=0.3,
|
||||
test_size = 0.2,
|
||||
random_state=None
|
||||
random_state=None,
|
||||
batch_size=64,
|
||||
learning_rate=0.00001,
|
||||
weight_decay=1e-5,
|
||||
lr_scheduler_patience=10,
|
||||
early_stop_patience=100,
|
||||
early_stop_threshold=0.99,
|
||||
):
|
||||
super(Qmlp, self).__init__(data=data, labels=labels, test_size=test_size, random_state=random_state)
|
||||
super(Qmlp, self).__init__(
|
||||
data=data,
|
||||
labels=labels,
|
||||
test_size=test_size,
|
||||
random_state=random_state,
|
||||
batch_size=batch_size,
|
||||
learning_rate=learning_rate,
|
||||
weight_decay=weight_decay,
|
||||
lr_scheduler_patience=lr_scheduler_patience,
|
||||
early_stop_patience=early_stop_patience,
|
||||
early_stop_threshold=early_stop_threshold,
|
||||
)
|
||||
|
||||
input_size = self.X_train.shape[1]
|
||||
num_classes = len(labels) if labels is not None else int(np.max(self.y_train)) + 1
|
||||
self.hidden_layers = hidden_layers
|
||||
self.dropout_rate = dropout_rate
|
||||
self.layers = nn.ModuleList()
|
||||
|
||||
|
||||
# 构造 MLP 网络结构
|
||||
self.build_model(input_shape=self.X_train.shape[1:], num_classes=self.num_classes)
|
||||
self._model_built = True
|
||||
|
||||
def build_model(self, input_shape, num_classes):
|
||||
if not self.hidden_layers:
|
||||
raise ValueError("'hidden_layers' must contain at least one layer size.")
|
||||
|
||||
input_size = int(np.prod(input_shape))
|
||||
self.layers = nn.ModuleList()
|
||||
|
||||
# 连接输入层和第一个隐藏层
|
||||
self.layers.append(nn.Linear(input_size, hidden_layers[0]))
|
||||
self.layers.append(nn.BatchNorm1d(hidden_layers[0]))
|
||||
self.layers.append(nn.Linear(input_size, self.hidden_layers[0]))
|
||||
self.layers.append(nn.BatchNorm1d(self.hidden_layers[0]))
|
||||
self.layers.append(nn.ReLU())
|
||||
self.layers.append(nn.Dropout(dropout_rate))
|
||||
|
||||
self.layers.append(nn.Dropout(self.dropout_rate))
|
||||
|
||||
# 创建隐藏层
|
||||
for i in range(1, len(hidden_layers)):
|
||||
self.layers.append(nn.Linear(hidden_layers[i-1], hidden_layers[i]))
|
||||
self.layers.append(nn.BatchNorm1d(hidden_layers[i]))
|
||||
self.layers.append(nn.ReLU())
|
||||
self.layers.append(nn.Dropout(dropout_rate))
|
||||
|
||||
for i in range(1, len(self.hidden_layers)):
|
||||
self.layers.append(nn.Linear(self.hidden_layers[i-1], self.hidden_layers[i]))
|
||||
self.layers.append(nn.BatchNorm1d(self.hidden_layers[i]))
|
||||
self.layers.append(nn.ReLU())
|
||||
self.layers.append(nn.Dropout(self.dropout_rate))
|
||||
|
||||
# 创建输出层
|
||||
self.layers.append(nn.Linear(hidden_layers[-1], num_classes))
|
||||
self.layers.append(nn.Linear(self.hidden_layers[-1], num_classes))
|
||||
self.__init_weights()
|
||||
|
||||
def forward(self, x):
|
||||
x = x.view(x.size(0), -1)
|
||||
for layer in self.layers:
|
||||
x = layer(x)
|
||||
x = layer(x)
|
||||
return x
|
||||
|
||||
def __init_weights(self):
|
||||
|
|
|
|||
|
|
@ -10,51 +10,85 @@ from Qfunctions.divSet import divSet as DS
|
|||
|
||||
|
||||
class Qnn(nn.Module):
|
||||
|
||||
def __init__(self, data, labels, test_size = 0.2, random_state=None):
|
||||
def __init__(
|
||||
self,
|
||||
data,
|
||||
labels,
|
||||
test_size=0.2,
|
||||
random_state=None,
|
||||
batch_size=64,
|
||||
learning_rate=0.00001,
|
||||
weight_decay=1e-5,
|
||||
lr_scheduler_patience=10,
|
||||
early_stop_patience=100,
|
||||
early_stop_threshold=0.99,
|
||||
):
|
||||
|
||||
super(Qnn, self).__init__()
|
||||
|
||||
# 使用gpu进行加速, 没有gpu的话使用CPU
|
||||
self.DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
||||
|
||||
|
||||
# 训练配置，子类共享
|
||||
self.batch_size = batch_size
|
||||
self.learning_rate = learning_rate
|
||||
self.weight_decay = weight_decay
|
||||
self.lr_scheduler_patience = lr_scheduler_patience
|
||||
self.early_stop_patience = early_stop_patience
|
||||
self.early_stop_threshold = early_stop_threshold
|
||||
|
||||
# 划分测试集和训练集
|
||||
self.X_train, self.X_test, self.y_train, self.y_test, self.LABEL_ENCODER = DS(
|
||||
data=data, labels=labels, test_size=test_size, random_state=random_state
|
||||
data=data, labels=labels, test_size=test_size, random_state=random_state
|
||||
)
|
||||
|
||||
|
||||
self.labels = labels
|
||||
|
||||
# 存储过程数据的文件
|
||||
self.epoch_data = {
|
||||
self.num_classes = len(labels) if labels is not None else int(np.max(self.y_train)) + 1
|
||||
|
||||
# 网络状态
|
||||
self._model_built = False
|
||||
|
||||
# 存储过程数据
|
||||
self.epoch_data = self._new_epoch_data()
|
||||
|
||||
# PCA 图片数据存储
|
||||
self.pca_2d, self.pca_3d = None, None
|
||||
|
||||
self.cm, self.cmn = None, None
|
||||
|
||||
def _new_epoch_data(self):
|
||||
return {
|
||||
'epoch': [],
|
||||
'train_loss': [],
|
||||
'train_accuracy': [],
|
||||
'test_accuracy': [],
|
||||
'precision': [],
|
||||
'recall': [],
|
||||
'recall': [],
|
||||
'f1_score': []
|
||||
}
|
||||
|
||||
# PCA 图片数据存储
|
||||
self.pca_2d, self.pca_3d = None, None
|
||||
|
||||
self.cm, self.cmn = None, None
|
||||
def build_model(self, input_shape, num_classes):
|
||||
# 子类必须实现具体网络结构
|
||||
raise NotImplementedError("Subclasses must implement build_model(input_shape, num_classes)")
|
||||
|
||||
def _transform_features(self, features):
|
||||
# 默认输入格式: [batch, feature_dim]
|
||||
return torch.tensor(features, dtype=torch.float32)
|
||||
|
||||
def _prepare_data(self):
|
||||
|
||||
# 将data转换为tensor形式
|
||||
X_train_tensor = torch.tensor(self.X_train, dtype=torch.float32)
|
||||
# 将data转换为tensor形式（子类可覆写 _transform_features）
|
||||
X_train_tensor = self._transform_features(self.X_train)
|
||||
y_train_tensor = torch.tensor(self.y_train, dtype=torch.long)
|
||||
|
||||
X_test_tensor = torch.tensor(self.X_test, dtype=torch.float32)
|
||||
X_test_tensor = self._transform_features(self.X_test)
|
||||
y_test_tensor = torch.tensor(self.y_test, dtype=torch.long)
|
||||
|
||||
train_dataset = TensorDataset(X_train_tensor, y_train_tensor)
|
||||
test_dataset = TensorDataset(X_test_tensor, y_test_tensor)
|
||||
|
||||
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
|
||||
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)
|
||||
train_loader = DataLoader(train_dataset, batch_size=self.batch_size, shuffle=True)
|
||||
test_loader = DataLoader(test_dataset, batch_size=self.batch_size, shuffle=False)
|
||||
|
||||
return train_loader, test_loader
|
||||
|
||||
|
|
@ -63,14 +97,15 @@ class Qnn(nn.Module):
|
|||
model = self.to(self.DEVICE)
|
||||
|
||||
criterion = nn.CrossEntropyLoss()
|
||||
optimizer = torch.optim.Adam(model.parameters(), lr=0.00001, weight_decay=1e-5)
|
||||
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=0.1, patience=10)
|
||||
optimizer = torch.optim.Adam(model.parameters(), lr=self.learning_rate, weight_decay=self.weight_decay)
|
||||
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
|
||||
optimizer,
|
||||
mode='min',
|
||||
factor=0.1,
|
||||
patience=self.lr_scheduler_patience,
|
||||
)
|
||||
best_test_accuracy = 0
|
||||
patience = 100
|
||||
counter = 0
|
||||
|
||||
# 99% 的准确率阈值
|
||||
accuracy_threshold = 0.99
|
||||
|
||||
for epoch in range(epochs_times):
|
||||
|
||||
|
|
@ -140,10 +175,10 @@ class Qnn(nn.Module):
|
|||
counter = 0
|
||||
else:
|
||||
counter += 1
|
||||
|
||||
if counter >= patience and best_test_accuracy >= accuracy_threshold:
|
||||
print(f"Early stopping at epoch {epoch+1}")
|
||||
break
|
||||
|
||||
if counter >= self.early_stop_patience and best_test_accuracy >= self.early_stop_threshold:
|
||||
print(f"Early stopping at epoch {epoch+1}")
|
||||
break
|
||||
|
||||
# cmn为归一化矩阵
|
||||
# Keep matrix dimensions stable even when some classes do not appear in this split.
|
||||
|
|
@ -155,6 +190,13 @@ class Qnn(nn.Module):
|
|||
return
|
||||
|
||||
def fit(self, epoch_times = 100):
|
||||
if not self._model_built:
|
||||
self.build_model(input_shape=self.X_train.shape[1:], num_classes=self.num_classes)
|
||||
self._model_built = True
|
||||
|
||||
# 每次训练前清空过程指标，避免重复累计
|
||||
self.epoch_data = self._new_epoch_data()
|
||||
|
||||
train_loader, test_loader = self._prepare_data()
|
||||
self._train_model(train_loader, test_loader, epochs_times=epoch_times)
|
||||
return
|
||||
|
|
@ -178,10 +220,12 @@ class Qnn(nn.Module):
|
|||
|
||||
# 外部获取混淆矩阵的接口
|
||||
def get_cm(self):
|
||||
return pd.DataFrame(self.cm, columns=self.labels, index=self.labels)
|
||||
label_names = self.labels if self.labels is not None else list(range(self.num_classes))
|
||||
return pd.DataFrame(self.cm, columns=label_names, index=label_names)
|
||||
|
||||
def get_cmn(self):
|
||||
return pd.DataFrame(self.cmn, columns=self.labels, index=self.labels)
|
||||
label_names = self.labels if self.labels is not None else list(range(self.num_classes))
|
||||
return pd.DataFrame(self.cmn, columns=label_names, index=label_names)
|
||||
|
||||
# 外部获取迭代数据的接口
|
||||
def get_epoch_data(self):
|
||||
|
|
|
|||
194
README.md
194
README.md
|
|
@ -1,29 +1,6 @@
|
|||
# Deeplearning 使用说明
|
||||
|
||||
## 1. 项目约定
|
||||
|
||||
### 1.1 输入数据格式
|
||||
每一类数据支持 `xls/xlsx/csv`。读取时默认取偶数列（索引 1,3,5...）作为特征，奇数列内容可忽略。
|
||||
|
||||
示意：
|
||||
|
||||
| 任意值 | 特征值 | 任意值 | 特征值 |
|
||||
|---|---|---|---|
|
||||
| arbitrary value | value | arbitrary value | value |
|
||||
|
||||
### 1.2 目录约定
|
||||
训练数据放在 `Static/`，输出结果放在 `Result/`。
|
||||
|
||||
推荐目录：
|
||||
|
||||
```text
|
||||
.
|
||||
├─ Static/
|
||||
│ └─ 20241009MaterialDiv/
|
||||
└─ Result/
|
||||
```
|
||||
|
||||
## 2. Conda 环境迁移
|
||||
## 1. Conda 环境迁移
|
||||
|
||||
环境文件在 `conda_env/`：
|
||||
|
||||
|
|
@ -31,7 +8,7 @@
|
|||
- `conda_env/environment.lock.txt`：精确锁定（同系统/同架构优先）
|
||||
- `conda_env/env.yml`：历史文件
|
||||
|
||||
### 2.1 创建环境
|
||||
### 1.1 创建环境
|
||||
|
||||
```bash
|
||||
# 方式1（推荐）：通用创建
|
||||
|
|
@ -47,7 +24,7 @@ python -V
|
|||
python -c "import torch; print(torch.__version__)"
|
||||
```
|
||||
|
||||
### 2.2 同名环境已存在时
|
||||
### 1.2 同名环境已存在时
|
||||
|
||||
```bash
|
||||
# 方式A：保留旧环境，改名创建
|
||||
|
|
@ -66,13 +43,86 @@ conda env create -f conda_env/environment.portable.yml
|
|||
conda activate Deeplearning
|
||||
```
|
||||
|
||||
### 2.3 重新导出环境
|
||||
### 1.3 重新导出环境
|
||||
|
||||
```bash
|
||||
conda env export -n Deeplearning --no-builds > conda_env/environment.portable.yml
|
||||
conda list -n Deeplearning --explicit > conda_env/environment.lock.txt
|
||||
```
|
||||
|
||||
### 1.4 主要依赖包
|
||||
|
||||
训练与数据处理核心依赖：
|
||||
|
||||
- Python 3.12
|
||||
- pytorch / torchvision / torchaudio
|
||||
- pandas / numpy / scipy
|
||||
- scikit-learn
|
||||
- matplotlib / seaborn
|
||||
- openpyxl / xlrd（用于 xls/xlsx 读写）
|
||||
|
||||
说明：项目仓库已提供完整环境文件，优先用 `conda_env/environment.portable.yml` 或 `conda_env/environment.lock.txt` 创建环境。
|
||||
|
||||
### 1.5 依赖安装方式（Conda / pip）
|
||||
|
||||
方式 A（推荐，Conda 一步到位）：
|
||||
|
||||
```bash
|
||||
conda env create -f conda_env/environment.portable.yml
|
||||
conda activate Deeplearning
|
||||
```
|
||||
|
||||
方式 B（Conda 最小安装，适合自定义环境）：
|
||||
|
||||
```bash
|
||||
conda create -n Deeplearning python=3.12 -y
|
||||
conda activate Deeplearning
|
||||
|
||||
# GPU 机器（CUDA 12.4）
|
||||
conda install -y pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia
|
||||
|
||||
# CPU 机器（无 CUDA）
|
||||
# conda install -y pytorch torchvision torchaudio cpuonly -c pytorch
|
||||
|
||||
conda install -y pandas numpy scipy scikit-learn matplotlib seaborn openpyxl xlrd
|
||||
```
|
||||
|
||||
方式 C（pip 安装，适合已有虚拟环境）：
|
||||
|
||||
```bash
|
||||
pip install torch torchvision torchaudio
|
||||
pip install pandas numpy scipy scikit-learn matplotlib seaborn openpyxl xlrd
|
||||
```
|
||||
|
||||
可选开发工具：
|
||||
|
||||
```bash
|
||||
pip install black autopep8 basedpyright
|
||||
```
|
||||
|
||||
## 2. 项目约定
|
||||
|
||||
### 2.1 输入数据格式
|
||||
每一类数据支持 `xls/xlsx/csv`。读取时默认取偶数列（索引 1,3,5...）作为特征，奇数列内容可忽略。
|
||||
|
||||
示意：
|
||||
|
||||
| 任意值 | 特征值 | 任意值 | 特征值 |
|
||||
|---|---|---|---|
|
||||
| arbitrary value | value | arbitrary value | value |
|
||||
|
||||
### 2.2 目录约定
|
||||
训练数据放在 `Static/`，输出结果放在 `Result/`。
|
||||
|
||||
推荐目录：
|
||||
|
||||
```text
|
||||
.
|
||||
├─ Static/
|
||||
│ └─ 20241009MaterialDiv/
|
||||
└─ Result/
|
||||
```
|
||||
|
||||
## 3. 快速开始
|
||||
|
||||
### 3.1 准备数据
|
||||
|
|
@ -137,44 +187,60 @@ PLA <-> PLA.xlsx 或 PLA/
|
|||
Wood <-> Wood.xlsx 或 Wood/
|
||||
```
|
||||
|
||||
### 3.3 训练示例
|
||||
### 3.3 通用：数据导入
|
||||
|
||||
```python
|
||||
from Qtorch.Models.Qmlp import Qmlp
|
||||
from Qfunctions.divSet import divSet
|
||||
from Qfunctions.loadData import load_data
|
||||
from Qfunctions.saveToXlsx import save_to_xlsx
|
||||
|
||||
projet_name = '20241009MaterialDiv'
|
||||
label_names = ['Acrlic', 'Ecoflex', 'PDMS', 'PLA', 'Wood']
|
||||
|
||||
# 自动识別数据模式
|
||||
# 支持 .xls 、.xlsx 、.csv 三种格式（可混合使用）
|
||||
# - folder/label.xlsx 或 folder/label.xls 或 folder/label.csv => 单文件模式
|
||||
# - folder/label/*.(xlsx|xls|csv) => 多子特征模式
|
||||
# 自动识别数据模式（支持 xls/xlsx/csv）
|
||||
data = load_data(projet_name, label_names)
|
||||
```
|
||||
|
||||
# 划分训练/测试集
|
||||
X_train, X_test, y_train, y_test, encoder = divSet(
|
||||
### 3.4 模型调用
|
||||
|
||||
#### 3.4.1 MLP
|
||||
|
||||
```python
|
||||
from Qtorch.Models.Qmlp import Qmlp
|
||||
|
||||
model = Qmlp(
|
||||
data=data,
|
||||
labels=label_names,
|
||||
test_size=0.3
|
||||
hidden_layers=[128, 256, 128],
|
||||
test_size=0.3,
|
||||
dropout_rate=0,
|
||||
)
|
||||
|
||||
# 构建模型
|
||||
model = Qmlp(
|
||||
X_train=X_train,
|
||||
X_test=X_test,
|
||||
y_train=y_train,
|
||||
y_test=y_test,
|
||||
hidden_layers=[128],
|
||||
dropout_rate=0
|
||||
)
|
||||
|
||||
# 训练与导出结果
|
||||
pca_2d, pca_3d = model.get_PCA()
|
||||
model.fit(300)
|
||||
```
|
||||
|
||||
#### 3.4.2 1D CNN
|
||||
|
||||
```python
|
||||
from Qtorch.Models.Qcnn import QCNN
|
||||
|
||||
model = QCNN(
|
||||
data=data,
|
||||
labels=label_names,
|
||||
conv_channels=(16, 32),
|
||||
kernel_size=3,
|
||||
hidden_size=128,
|
||||
test_size=0.3,
|
||||
dropout_rate=0,
|
||||
)
|
||||
|
||||
model.fit(300)
|
||||
```
|
||||
|
||||
### 3.5 通用：结果获取与图表导出
|
||||
|
||||
```python
|
||||
from Qfunctions.saveToXlsx import save_to_xlsx
|
||||
|
||||
pca_2d, pca_3d = model.get_PCA()
|
||||
cm = model.get_cm()
|
||||
cmn = model.get_cmn()
|
||||
epoch_data = model.get_epoch_data()
|
||||
|
|
@ -196,7 +262,9 @@ save_to_xlsx(project_name=projet_name, file_name='acc_and_loss', data=epoch_data
|
|||
自动识别规则：
|
||||
|
||||
- 若每个 `label` 都对应 `folder/label/*.(xlsx|xls|csv)`，识别为多子特征模式。
|
||||
- 若每个 `label` 都对应 `folder/label.(xlsx|xls|csv)`，识别为单文件模式。- 超出需法的文件格式（只许 xls/xlsx/csv），汽转时报错。- 若两种都成立（同名文件和同名子目录同时存在），会报错并提示只保留一种目录结构。
|
||||
- 若每个 `label` 都对应 `folder/label.(xlsx|xls|csv)`，识别为单文件模式。
|
||||
- 超出支持范围的文件格式（仅支持 xls/xlsx/csv）会报错。
|
||||
- 若两种都成立（同名文件和同名子目录同时存在），会报错并提示只保留一种目录结构。
|
||||
- 若两种都不成立，会报错并提示检查目录结构或 `label_names`。
|
||||
|
||||
读取路径规则：
|
||||
|
|
@ -211,3 +279,27 @@ save_to_xlsx(project_name=projet_name, file_name='acc_and_loss', data=epoch_data
|
|||
|
||||
- `label_names` 与文件/文件夹是否同名
|
||||
- 文件后缀是否为 `.xls`、`.xlsx` 或 `.csv`（其他格式将报错）
|
||||
|
||||
## 6. TODO（后续计划）
|
||||
|
||||
### 阶段一：基础稳定与可维护性
|
||||
|
||||
- [ ] 固化模型基类契约：`build_model`、输入变换钩子、统一训练配置。
|
||||
- [ ] 封装 `QDL` 包结构（`qdl.data / qdl.models / qdl.export / qdl.api`）。
|
||||
- [ ] 增加兼容层：保留旧导入路径，逐步迁移到新包路径。
|
||||
- [ ] 增加最小测试集：`load_data`、`Qmlp.fit(1)`、`QCNN.fit(1)`、导出函数。
|
||||
|
||||
### 阶段二：高维与复合模型能力
|
||||
|
||||
- [ ] 将 batch 结构升级为字典（`x/y/lengths/mask/meta`），支持复合输入。
|
||||
- [ ] 在训练框架中加入钩子：`prepare_batch`、`forward_step`、`compute_loss`。
|
||||
- [ ] 引入编码器层（Encoder）抽象：按策略扩展（MLP/CNN/LSTM/GNN/Transformer），不按维度硬扩类。
|
||||
- [ ] 支持可变长时间序列（`collate_fn + lengths + mask`），作为后续复合模型基础。
|
||||
- [ ] 增加多分支融合模板（时序分支 + 静态分支），预留多任务损失组合。
|
||||
|
||||
### 阶段三：对外能力与发布
|
||||
|
||||
- [ ] 统一对外入口：提供高层 API（例如 `train_mlp`、`train_cnn1d`、`train_hybrid`）。
|
||||
- [ ] 在模型文档中预留扩展位：`3.4.3 LSTM`、`3.4.4 GNN`、`3.4.5 Hybrid`。
|
||||
- [ ] 完成打包配置（`pyproject.toml`）与本地可编辑安装说明。
|
||||
- [ ] 发布前回归：在 Conda 与 pip 环境分别跑通最小端到端流程。
|
||||
|
|
|
|||
Loading…
Reference in New Issue