[Pytorch][转载]用pytorch解决简单的FizzBuzz问题

FizzBuzz FizzBuzz是一个简单的小游戏。游戏规则如下：从1开始往上数数，当遇到3的倍数的时候，说fizz，当遇到5的倍数，说buzz，当遇到15的倍数，就说fizzbuzz，其他情况下则正常数数。   我们可以写一个简单的小程序来决定要返回正常数值还是fizz, buzz 或者 fizzbuzz。  # One-hot encode the desired outputs: [number, "fizz", "buzz", "fizzbuzz"] def fizz_buzz_encode(i):     if   i % 15 == 0: return 3     elif i % 5  == 0: return 2     elif i % 3  == 0: return 1     else:             return 0      def fizz_buzz_decode(i, prediction):     return [str(i), "fizz", "buzz", "fizzbuzz"][prediction]   print(fizz_buzz_decode(1, fizz_buzz_encode(1))) print(fizz_buzz_decode(2, fizz_buzz_encode(2))) print(fizz_buzz_decode(5, fizz_buzz_encode(5))) print(fizz_buzz_decode(12, fizz_buzz_encode(12))) print(fizz_buzz_decode(15, fizz_buzz_encode(15))) import numpy as np import torch   NUM_DIGITS = 10   # Represent each input by an array of its binary digits. def binary_encode(i, num_digits):     return np.array([i >> d & 1 for d in range(num_digits)])   trX = torch.Tensor([binary_encode(i, NUM_DIGITS) for i in range(101, 2 ** NUM_DIGITS)]) trY = torch.LongTensor([fizz_buzz_encode(i) for i in range(101, 2 ** NUM_DIGITS)])    然后我们用PyTorch定义模型  # Define the model NUM_HIDDEN = 100 model = torch.nn.Sequential(     torch.nn.Linear(NUM_DIGITS, NUM_HIDDEN),     torch.nn.ReLU(),     torch.nn.Linear(NUM_HIDDEN, 4) ) 为了让我们的模型学会FizzBuzz这个游戏，我们需要定义一个损失函数，和一个优化算法。 这个优化算法会不断优化（降低）损失函数，使得模型的在该任务上取得尽可能低的损失值。 损失值低往往表示我们的模型表现好，损失值高表示我们的模型表现差。 由于FizzBuzz游戏本质上是一个分类问题，我们选用Cross Entropyy Loss函数。 优化函数我们选用Stochastic Gradient Descent。  loss_fn = torch.nn.CrossEntropyLoss() optimizer = torch.optim.SGD(model.parameters(), lr = 0.05)    以下是模型的训练代码  # Start training it BATCH_SIZE = 128 for epoch in range(10000):     for start in range(0, len(trX), BATCH_SIZE):         end = start + BATCH_SIZE         batchX = trX[start:end]         batchY = trY[start:end]           y_pred = model(batchX)         loss = loss_fn(y_pred, batchY)           optimizer.zero_grad()         loss.backward()         optimizer.step()       # Find loss on training data     loss = loss_fn(model(trX), trY).item()     print('Epoch:', epoch, 'Loss:', loss)  最后我们用训练好的模型尝试在1到100这些数字上玩FizzBuzz游戏  # Output now testX = torch.Tensor([binary_encode(i, NUM_DIGITS) for i in range(1, 101)]) with torch.no_grad():     testY = model(testX) predictions = zip(range(1, 101), list(testY.max(1)[1].data.tolist()))   print([fizz_buzz_decode(i, x) for (i, x) in predictions])  print(np.sum(testY.max(1)[1].numpy() == np.array([fizz_buzz_encode(i) for i in range(1,101)]))) testY.max(1)[1].numpy() == np.array([fizz_buzz_encode(i) for i in range(1,101)])