第五课第二周笔记

1. one-hot

Onehot转换

def convert_to_one_hot(Y, C):
    Y = np.eye(C)[Y.reshape(-1)]
    return Y

2. 词和index以及词向量的转换

def read_glove_vecs(glove_file):
    with open(glove_file, 'r', encoding='utf8') as f:
        words = set()  # 集合避免重复
        word_to_vec_map = {}  # 词到词向量
        for line in f:  # 默认按行读取
            line = line.strip().split()  # strip移除字符串头尾指定的字符序列,默认为空格
            curr_word = line[0]  # 取出单词
            words.add(curr_word)  # 添加单词
            word_to_vec_map[curr_word] = np.array(line[1:], dtype=np.float64)  # 用字典建立索引
        i = 1
        words_to_index = {}
        index_to_words = {}
        for w in sorted(words):  # sorted可以对所有可迭代对象进行排序操作，返回一个新的list
            words_to_index[w] = i
            index_to_words[i] = w
            i += 1
        return words_to_index, index_to_words, word_to_vec_map

3. 加载训练好的词向量

Embedding层

1. 经过下面函数，完成了I love you ---> [185457, 2262788, 394475, 0, 0]的转换

def sentences_to_indices(X, word_to_index, max_len):
    m = X.shape[0]  # 有多少个句子
    x_indices = np.zeros(shape=(m, max_len))
    for i in range(m):
        sentence_words = (X[i].lower()).split()
        j = 0
        for w in sentence_words:
            X_indices[i, j] = word_to_index[w]  # 将每句话中的单词索引为index
            j +=1
    return x_indices

1. 加载之前训练好的词向量

   def pretrained_embedding_layer(word_to_vec_map, word_to_index):
      vocab_len = len(word_to_index) + 1  # 词的数量
      emb_dim = word_to_vec_map["cucumber"].shape[0]  # 单个词向量的维度=50
      embedding_layer = Embedding(input_dim=vocab_len, out_dim=emb_dim, trainable=False)  # input_dim词汇表大小，即最大整数Index+1  out_dim:词向量维度
      # 设置权重
      emb_matrix = np.zeros(shape=(vocab_len, emb_dim))
      for word, index in word_to_index.items():  # 循环单词和索引
          emb_matrix[index, :] = word_to_vec_map[word]
      # 设置嵌入层权重之前，需要build嵌入层
      embedding_layer.build((None,))
      embedding_layer.set_weights([emb_matrix])
      return embedding_layer

4 第一个模型

def Emjify_V2(input_shape, word_to_vec_map, word_to_index):
    """
    input_shape -- shape of the input, usually (max_len,)
    """
    sentence_indices = Input(shape=input_shape, dtype='int32')
    embedding_layer = pretrained_embedding_layer(word_to_vec_map, word_to_index)
    embeddings = emedding_layer(sentence_indices)  # 自定义的Layer对象的方法,第二个括号可以传递入张量
    X = LSTM(units=128, return_sequence=True)(embeddings)  # return_sequence是否全部接受
    X = Dropout(0.5)(X)
    X = LSTM(units=128, return_sequences=False)(X)
    X = Dropout(0.5)(X)
    X = Dense(units=5)(X)
    X = Activation('softmax')(X)
    model = Model(inputs=sentence_indices, outputs=X)
    return model