【Pytorch】Fizz Buzz

学习参考来自：

• Fizz Buzz in Tensorflow
• https://github.com/wmn7/ML_Practice/tree/master/2019_06_10
• Fizz Buzz in Pytorch

---

I need you to print the numbers from 1 to 100, except that if the number is divisible by 3 print “fizz”, if it’s divisible by 5 print “buzz”, and if it’s divisible by 15 print “fizzbuzz”.

编程题很简单，我们用 MLP 实现试试

思路，训练集数据101~1024，对其进行某种规则的编码，标签为经分类 one-hot 编码后的标签
测试集，1~100

don’t say so much, show me the code.

1 数据编码

import numpy as np
import torch
import torch.nn as nn
import torch.utils.data as Data

def binary_encode(i, num_digits):
    """将每个input转换为binary digits(转换为二进制的表示, 最多可是表示2^num_digits)
    :param i:
    :param num_digits:
    :return:
    """
    return np.array([i >> d & 1 for d in range(num_digits)])

编码形式，依次除以 $2^{0,1,2,3,...}$，结果按位与 1

m & 1，结果为 0 表示 m 为偶数， 结果为 1 表示 m 为奇数

$>>m$ 右移表示除以 $2^m$

第一位就能表示奇偶了，所有数字编码都不一样

eg，101 进行 num_digits=10 编码后结果为 1 0 1 0 0 1 1 0 0 0

步骤

101 / 1 = 101 奇数 1
101 / 2 = 50 偶数 0
101 / 4 = 25 奇数 1
101 / 8 = 12 偶数 0
101 / 16 = 6 偶数 0
101 / 32 = 3 奇数 1
101 / 64 = 1 奇数 1
101 / 128 = 0 偶数 0
101 / 256= 0 偶数 0
101 / 512= 0 偶数 0

标签，0，1，2，3 四个类别

def fizz_buzz_encode(i):
    """将output转换为lebel
    :param i:
    :return:
    """
    if i % 15 == 0:  # fizzbuzz
        return 3
    elif i % 5 == 0:  # buzz
        return 2
    elif i % 3 == 0:  # fizz
        return 1
    else:
        return 0

编码长度设定，数据集 101 ~ 1024

NUM_DIGITS = 10
trX = np.array([binary_encode(i, NUM_DIGITS) for i in range(101, 2**NUM_DIGITS)])  # 101~1024
trY = np.array([fizz_buzz_encode(i) for i in range(101, 2**NUM_DIGITS)])

# print(len(trX), len(trY))  # 923 923
# print(trX[:5])
"""
[[1 0 1 0 0 1 1 0 0 0]
 [0 1 1 0 0 1 1 0 0 0]
 [1 1 1 0 0 1 1 0 0 0]
 [0 0 0 1 0 1 1 0 0 0]
 [1 0 0 1 0 1 1 0 0 0]]
"""
# print(trY[:5])  # [0 1 0 0 3]

2 网络搭建

搭建简单的 MLP 网络

class FizzBuzzModel(nn.Module):
    def __init__(self, in_features, out_classes, hidden_size, n_hidden_layers):
        super(FizzBuzzModel,self).__init__()
        layers = []
        for i in range(n_hidden_layers):
            layers.append(nn.Linear(hidden_size,hidden_size))
            # layers.append(nn.Dropout(0.5))
            layers.append(nn.BatchNorm1d(hidden_size))
            layers.append(nn.ReLU())
        self.inputLayer = nn.Linear(in_features, hidden_size)
        self.relu = nn.ReLU()
        self.layers = nn.Sequential(*layers)  # 重复的搭建隐藏层
        self.outputLayer = nn.Linear(hidden_size, out_classes)

    def forward(self, x):
        x = self.inputLayer(x)
        x = self.relu(x)
        x = self.layers(x)
        out = self.outputLayer(x)
        return out

初始化网络，看看网络结构

# Device configuration
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# define the model
simpleModel = FizzBuzzModel(NUM_DIGITS, 4, 150, 3).to(device)
print(simpleModel)
"""
FizzBuzzModel(
  (inputLayer): Linear(in_features=10, out_features=150, bias=True)
  (relu): ReLU()
  (layers): Sequential(
    (0): Linear(in_features=150, out_features=150, bias=True)
    (1): BatchNorm1d(150, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
    (3): Linear(in_features=150, out_features=150, bias=True)
    (4): BatchNorm1d(150, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (5): ReLU()
    (6): Linear(in_features=150, out_features=150, bias=True)
    (7): BatchNorm1d(150, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (8): ReLU()
  )
  (outputLayer): Linear(in_features=150, out_features=4, bias=True)
)
"""

输入 10， 输出4，隐藏层维度 150，隐藏层重复了 3 次

3 网络配置，训练

定义下超参数，损失函数，优化器，载入数据训练，输出训练精度与损失

# Loss and optimizer
learning_rate = 0.05
criterion = nn.CrossEntropyLoss()
# optimizer = torch.optim.Adam(simpleModel.parameters(), lr=learning_rate)
optimizer = torch.optim.SGD(simpleModel.parameters(), lr=learning_rate)

# 使用batch进行训练
FizzBuzzDataset = Data.TensorDataset(torch.from_numpy(trX).float().to(device),
                                     torch.from_numpy(trY).long().to(device))

loader = Data.DataLoader(dataset=FizzBuzzDataset,
                         batch_size=128*5,
                         shuffle=True)

# 进行训练
simpleModel.train()
epochs = 3000

for epoch in range(1, epochs):
    for step, (batch_x, batch_y) in enumerate(loader):
        out = simpleModel(batch_x)  # 前向传播
        loss = criterion(out, batch_y)  # 计算损失
        optimizer.zero_grad()  # 梯度清零
        loss.backward()  # 反向传播
        optimizer.step()  # 随机梯度下降
    correct = 0
    total = 0
    _, predicted = torch.max(out.data, 1)
    total += batch_y.size(0)
    correct += (predicted == batch_y).sum().item()
    acc = 100*correct/total
    print('Epoch : {:0>4d} | Loss : {:<6.4f} | Train Accuracy : {:<6.2f}%'.format(epoch, loss, acc))

"""
Epoch : 0001 | Loss : 1.5343 | Train Accuracy : 14.63 %
Epoch : 0002 | Loss : 1.9779 | Train Accuracy : 42.58 %
Epoch : 0003 | Loss : 2.4198 | Train Accuracy : 53.41 %
Epoch : 0004 | Loss : 1.7360 | Train Accuracy : 53.41 %
Epoch : 0005 | Loss : 1.3161 | Train Accuracy : 49.73 %
Epoch : 0006 | Loss : 1.4866 | Train Accuracy : 22.75 %
Epoch : 0007 | Loss : 1.3993 | Train Accuracy : 25.57 %
Epoch : 0008 | Loss : 1.2428 | Train Accuracy : 28.49 %
Epoch : 0009 | Loss : 1.1906 | Train Accuracy : 44.31 %
Epoch : 0010 | Loss : 1.1929 | Train Accuracy : 52.44 %
...
Epoch : 2990 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2991 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2992 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2993 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2994 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2995 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2996 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2997 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2998 | Loss : 0.0000 | Train Accuracy : 100.00%
Epoch : 2999 | Loss : 0.0000 | Train Accuracy : 100.00%
"""

训练集上精度是 OK 的，能到 100%，下面看看测试集上的精度

4 结果预测

把 one-hot 标签转化成 fizz buzz 的形式

def fizz_buzz_decode(i, prediction):
    return [str(i), "fizz", "buzz", "fizzbuzz"][prediction]

载入测试集，开始预测

simpleModel.eval()
# 进行预测
testX = np.array([binary_encode(i, NUM_DIGITS) for i in range(1, 101)])
predicts = simpleModel(torch.from_numpy(testX).float().to(device))
# 预测的结果
_, res = torch.max(predicts, 1)
print(res)
"""
tensor([0, 0, 0, 1, 0, 0, 0, 2, 1, 0, 1, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 3, 1, 0,
        0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        1, 1, 1, 1, 2, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 1, 0, 1, 1, 1, 0,
        0, 0, 0, 0], device='cuda:0')
"""

# 格式的转换
predictions = [fizz_buzz_decode(i, prediction) for (i, prediction) in zip(range(1, 101), res)]
print(predictions)
"""
['1', '2', '3', 'fizz', '5', '6', '7', 'buzz', 'fizz', '10', 'fizz', 'fizzbuzz', 'fizzbuzz', 'fizz', 'fizz', '16', '17', '18', '19', '20', '21', 'fizzbuzz', 'fizz', '24', '25', '26', '27', '28', '29', '30', 'fizz', '32', '33', '34', '35', '36', '37', '38', '39', 'fizz', '41', 'fizz', '43', '44', '45', 'fizz', '47', '48', '49', '50', '51', '52', '53', '54', '55', '56', '57', '58', '59', '60', '61', 'fizz', '63', '64', '65', '66', '67', '68', '69', '70', '71', '72', 'fizz', 'fizz', 'fizz', 'fizz', 'buzz', 'buzz', 'fizz', '80', '81', '82', '83', '84', '85', '86', '87', 'fizzbuzz', '89', '90', 'fizz', '92', 'fizz', 'fizz', 'fizz', '96', '97', '98', '99', '100']
"""

5 翻车现场

对比下标签

labels = []
for i in range(1, 101):
    if i % 15 == 0:  # fizzbuzz
        labels.append("fizzbuzz")
    elif i % 5 == 0:  # buzz
        labels.append("buzz")
    elif i % 3 == 0:  # fizz
        labels.append("fizz")
    else:
        labels.append(str(i))
print(labels)
print(labels == predictions)

"""
['1', '2', 'fizz', '4', 'buzz', 'fizz', '7', '8', 'fizz', 'buzz', '11', 'fizz', '13', '14', 'fizzbuzz', '16', '17', 'fizz', '19', 'buzz', 'fizz', '22', '23', 'fizz', 'buzz', '26', 'fizz', '28', '29', 'fizzbuzz', '31', '32', 'fizz', '34', 'buzz', 'fizz', '37', '38', 'fizz', 'buzz', '41', 'fizz', '43', '44', 'fizzbuzz', '46', '47', 'fizz', '49', 'buzz', 'fizz', '52', '53', 'fizz', 'buzz', '56', 'fizz', '58', '59', 'fizzbuzz', '61', '62', 'fizz', '64', 'buzz', 'fizz', '67', '68', 'fizz', 'buzz', '71', 'fizz', '73', '74', 'fizzbuzz', '76', '77', 'fizz', '79', 'buzz', 'fizz', '82', '83', 'fizz', 'buzz', '86', 'fizz', '88', '89', 'fizzbuzz', '91', '92', 'fizz', '94', 'buzz', 'fizz', '97', '98', 'fizz', 'buzz']
False
"""

哈哈哈， False 翻车了，尝试了很多次，很难 True