《动手学深度学习》笔记①：SoftMax和RNN基础

李沐的经典著作《动手学深度学习》，推荐伯禹AI动手深度学习。

此系列为个人的学习和补充笔记，方便日后回顾。

一、SoftMax

推荐下面这篇博文，详细总结SoftMax。

详解softmax函数以及相关求导过程

二、RNN基础

用Pytorch版TextRNN，实现predict next step

import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable

dtype = torch.FloatTensor

sentences = [ "i like dog", "i love coffee", "i hate milk"]

word_list = " ".join(sentences).split()
word_list = list(set(word_list))  # 去重
print(word_list)

word_dict = {w: i for i, w in enumerate(word_list)}
number_dict = {i: w for i, w in enumerate(word_list)}
print(word_dict)
print(number_dict)

n_class = len(word_dict)  # 词表大小，做n分类预测

# TextRNN Parameter
batch_size = len(sentences)
n_step = 2 # number of cells(= number of Step)
n_hidden = 5 # number of hidden units in one cell

def make_batch(sentences):
    input_batch = []
    target_batch = []

    for sen in sentences:
        word = sen.split()
        input = [word_dict[n] for n in word[:-1]]  
        target = word_dict[word[-1]]  # 最后一个单词作为target
        print("input:", input, "traget:", target)
        
        input_batch.append(np.eye(n_class)[input])
        target_batch.append(target)

    return input_batch, target_batch

# to Torch.Tensor
input_batch, target_batch = make_batch(sentences)

input_batch = Variable(torch.Tensor(input_batch))
target_batch = Variable(torch.LongTensor(target_batch))

print(input_batch)
print(target_batch)  # [5 : dog, 0 : coffee, 1 : milk]

class TextRNN(nn.Module):
    def __init__(self):
        super(TextRNN, self).__init__()

        self.rnn = nn.RNN(input_size=n_class, hidden_size=n_hidden)
        self.W = nn.Parameter(torch.randn([n_hidden, n_class]).type(dtype))
        self.b = nn.Parameter(torch.randn([n_class]).type(dtype))

    def forward(self, hidden, X):
        X = X.transpose(0, 1) # X : [n_step, batch_size, n_class]  [2, 3, 7]
        outputs, hidden = self.rnn(X, hidden)
        # outputs : [n_step, batch_size, num_directions(=1) * n_hidden]
        # hidden : [num_layers(=1) * num_directions(=1), batch_size, n_hidden]
        print("X : ",X ,"outputs : ",outputs ,"hidden : ",hidden)
        
        outputs = outputs[-1] # [batch_size, num_directions(=1) * n_hidden]
        model = torch.mm(outputs, self.W) + self.b # model : [batch_size, n_class]
        return model


model = TextRNN()

criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)

# Training
for epoch in range(5000):
    optimizer.zero_grad()

    # hidden : [num_layers * num_directions, batch, hidden_size]
    hidden = Variable(torch.zeros(1, batch_size, n_hidden))
    # input_batch : [batch_size, n_step, n_class]
    output = model(hidden, input_batch)

    # output : [batch_size, n_class], target_batch : [batch_size] (LongTensor, not one-hot)
    loss = criterion(output, target_batch)
    if (epoch + 1) % 1000 == 0:
        print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))

    loss.backward()
    optimizer.step()

Epoch: 1000 cost = 0.064772

Epoch: 2000 cost = 0.014707

Epoch: 3000 cost = 0.005852

Epoch: 4000 cost = 0.002837

Epoch: 5000 cost = 0.001509

input = [sen.split()[:2] for sen in sentences]
print(input)

# Predict
hidden = Variable(torch.zeros(1, batch_size, n_hidden))
predict = model(hidden, input_batch).data.max(1, keepdim=True)[1]
print([sen.split()[:2] for sen in sentences], '->', [number_dict[n.item()] for n in predict.squeeze()])