CSE5001_Project

A Robust Handwritten Digit Recognition Network.

Start Up

First create a conda virtual environment with the following script.

CONDA_RESTORE_FREE_CHANNEL=1 conda env create -f environment.yml

or

conda env create -f environment4mac.yml

Develop

Step 1

Create or reuse a net work in ./network.

Step 2

Implement your algorithm in ./alg, with the base class class BaseAlg(ABC):.

class BaseAlg(ABC):
	@abstractmethod
	def train(self, train_loader):
		"""
		训练算法
		:param train_loader: 训练数据加载器
		"""
		raise NotImplementedError

	@abstractmethod
	def predict(self, data_loader):
		"""
		使用训练好的算法进行预测
		:param test_loader: 测试数据加载器
		:return: 预测结果
		"""
		raise NotImplementedError

	@abstractmethod
	def evaluate(self, test_loader):
		"""
		评估算法性能
		:param test_loader: 测试数据加载器
		:return: 算法性能指标
		"""
		raise NotImplementedError
	
	@abstractmethod
	def save(self, path):
		"""
		保存模型文件
		:param path: 保存路径
		"""
		raise NotImplementedError
	
	@abstractmethod
	def load(self, path):
		"""
		将模型文件导入
		:param path: 保存路径
		"""
		raise NotImplementedError

Step 3

Evaluate your algorithm in main.ipynb.

TODO

• Implement visualized components.

• Try different algorithms.