元学习 - MAML python实现完整代码

import torch
from torch import nn, optim
import numpy as np

# 数据集生成器
def generate_data(num_tasks=10):
    dataset = []
    for _ in range(num_tasks):
        x = torch.from_numpy(np.random.rand(5, 1).astype(np.float32))  # 生成随机数据
        y = torch.from_numpy(np.random.rand(5, 1).astype(np.float32))
        dataset.append((x, y))
    return dataset

# MAML算法
class MAML:
    def __init__(self, model):
        self.model = model
        self.meta_optimizer = optim.Adam(model.parameters(), lr=0.001)  # 元优化器

    def inner_update(self, model, data, lr_inner=0.01, num_updates=1):
        optimizer = optim.SGD(model.parameters(), lr=lr_inner)
        x, y = data

        for _ in range(num_updates):
            loss = nn.MSELoss()(model(x), y)  # 损失计算
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

    def meta_update(self, tasks, iterations=100, num_updates=1):
        for _ in range(iterations):
            for task in tasks:
                data = task
                cloned_model = self.clone_model()  # 克隆模型
                self.inner_update(cloned_model, data, lr_inner=0.01, num_updates=num_updates)  # 内部更新

                for p, cloned_p in zip(self.model.parameters(), cloned_model.parameters()):
                    # 更新原始模型的参数
                    if p.grad is None:
                        p.grad = cloned_p.grad.clone()
                    else:
                        p.grad += cloned_p.grad.clone()
                self.meta_optimizer.step()

    def clone_model(self):
        # 克隆模型的参数
        model_clone = SimpleModel()
        model_clone.load_state_dict(self.model.state_dict())
        return model_clone

# 简单的模型
class SimpleModel(nn.Module):
    def __init__(self):
        super(SimpleModel, self).__init__()
        self.fc = nn.Linear(1, 1)

    def forward(self, x):
        return self.fc(x)

# 生成数据集
tasks = generate_data(num_tasks=10)

# 初始化MAML和模型
model = SimpleModel()
maml = MAML(model)

# 进行元学习
maml.meta_update(tasks, iterations=100, num_updates=1)
 

 

import torch
from torch import nn, optim
import numpy as np

# 数据集生成器
def generate_data(num_tasks=10):
    dataset = []
    for _ in range(num_tasks):
        x = torch.from_numpy(np.random.rand(5, 1).astype(np.float32))  # 生成随机数据
        y = torch.from_numpy(np.random.rand(5, 1).astype(np.float32))
        dataset.append((x, y))
    return dataset

# MAML算法
class MAML:
    def __init__(self, model):
        self.model = model
        self.meta_optimizer = optim.Adam(model.parameters(), lr=0.001)  # 元优化器

    def inner_update(self, model, data, lr_inner=0.01, num_updates=1):
        optimizer = optim.SGD(model.parameters(), lr=lr_inner)
        x, y = data

        for _ in range(num_updates):
            loss = nn.MSELoss()(model(x), y)  # 损失计算
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

    def meta_update(self, tasks, iterations=100, num_updates=1):
        for _ in range(iterations):
            for task in tasks:
                data = task
                cloned_model = self.clone_model()  # 克隆模型
                self.inner_update(cloned_model, data, lr_inner=0.01, num_updates=num_updates)  # 内部更新

                for p, cloned_p in zip(self.model.parameters(), cloned_model.parameters()):
                    # 更新原始模型的参数
                    if p.grad is None:
                        p.grad = cloned_p.grad.clone()
                    else:
                        p.grad += cloned_p.grad.clone()
                self.meta_optimizer.step()

    def clone_model(self):
        # 克隆模型的参数
        model_clone = SimpleModel()
        model_clone.load_state_dict(self.model.state_dict())
        return model_clone

# 简单的模型
class SimpleModel(nn.Module):
    def __init__(self):
        super(SimpleModel, self).__init__()
        self.fc = nn.Linear(1, 1)

    def forward(self, x):
        return self.fc(x)

# 生成数据集
tasks = generate_data(num_tasks=10)

# 初始化MAML和模型
model = SimpleModel()
maml = MAML(model)

# 进行元学习
maml.meta_update(tasks, iterations=100, num_updates=1)