Sequential Decision Tree Model

This project implements a Sequential Decision Tree model for image classification tasks, with support for CIFAR-10 and CIFAR-100 datasets.

English | 中文

English

Key Features

• Flexible model architecture supporting various backbone networks (ResNet, ConvMixer, RDNet, etc.)
• Support for both CIFAR-10 and CIFAR-100 datasets
• Distributed Data Parallel (DDP) training support for multi-GPU setups

Configuration

All main parameters can be adjusted in the Paper_global_vars.py file. This includes settings such as:

• Number of epochs
• Input size
• Learning rate
• Batch size
• Dataset selection
• Model selection
• Optimizer choice
• Model save path

Running the Training

Single GPU

For single GPU training, simply run:

python Paper_trainForMultiDDP.py

Multi-GPU

For multi-GPU training using DDP, use the following command:

python -m torch.distributed.launch --nproc_per_node=<num_gpus> Paper_trainForMultiDDP.py

Replace <num_gpus> with the number of GPUs you want to use.

Core Components

The core of this project is the Paper_Tree.py file, which integrates multiple model architectures:

• SequentialDecisionTree (for CIFAR-10)
• SequentialDecisionTreeCIFAR100 (for CIFAR-100)
• SequentialDecisionTreeForRDNet (RDNet-based for CIFAR-10)
• SequentialDecisionTreeCIFAR100ForRDNet (RDNet-based for CIFAR-100)

These models incorporate various backbone networks such as ResNet, ConvMixer, and RDNet. You can easily integrate your own custom models and experiment with different architectures.

Customization

To use your own models or datasets:

1. Add your model implementation to the Paper_Tree.py file.
2. Update the global_vars in Paper_global_vars.py to include your new model.
3. Modify the data loading process in Paper_trainForMultiDDP.py if using a custom dataset.

Feel free to experiment with different model architectures and training configurations to achieve the best results for your specific use case.

---

中文

主要特点

• 灵活的模型架构，支持各种骨干网络（ResNet、ConvMixer、RDNet等）
• 支持CIFAR-10和CIFAR-100数据集
• 支持多GPU设置的分布式数据并行（DDP）训练

配置

所有主要参数都可以在Paper_global_vars.py文件中调整。这包括以下设置：

• 训练轮数
• 输入大小
• 学习率
• 批量大小
• 数据集选择
• 模型选择
• 优化器选择
• 模型保存路径

运行训练

单GPU

对于单GPU训练，只需运行：

python Paper_trainForMultiDDP.py

多GPU

对于使用DDP的多GPU训练，使用以下命令：

python -m torch.distributed.launch --nproc_per_node=<num_gpus> Paper_trainForMultiDDP.py

将<num_gpus>替换为您想使用的GPU数量。

核心组件

本项目的核心是Paper_Tree.py文件，它集成了多个模型架构：

• SequentialDecisionTree（用于CIFAR-10）
• SequentialDecisionTreeCIFAR100（用于CIFAR-100）
• SequentialDecisionTreeForRDNet（基于RDNet的CIFAR-10）
• SequentialDecisionTreeCIFAR100ForRDNet（基于RDNet的CIFAR-100）

这些模型包含了各种骨干网络，如ResNet、ConvMixer和RDNet。您可以轻松集成自己的自定义模型并尝试不同的架构。

自定义

要使用您自己的模型或数据集：

1. 将您的模型实现添加到Paper_Tree.py文件中。
2. 更新Paper_global_vars.py中的global_vars以包含您的新模型。
3. 如果使用自定义数据集，请修改Paper_trainForMultiDDP.py中的数据加载过程。

请随意尝试不同的模型架构和训练配置，以达到您特定用例的最佳结果。