python - 如何合并数值和嵌入序列模型以处理 RNN 中的类别

Question

我想为我的分类特征构建一个带有嵌入的单层 LSTM 模型。我目前有数字特征和一些分类特征，比如位置，它们不能被单热编码，例如由于计算复杂性，使用 pd.get_dummies()，这是我最初打算做的。

让我们想象一个例子:

示例数据

data = {
    'user_id': [1,1,1,1,2,2,3],
    'time_on_page': [10,20,30,20,15,10,40],
    'location': ['London','New York', 'London', 'New York', 'Hong Kong', 'Tokyo', 'Madrid'],
    'page_id': [5,4,2,1,6,8,2]
}
d = pd.DataFrame(data=data)
print(d)
   user_id  time_on_page   location  page_id
0        1            10     London        5
1        1            20   New York        4
2        1            30     London        2
3        1            20   New York        1
4        2            15  Hong Kong        6
5        2            10      Tokyo        8
6        3            40     Madrid        2

让我们看看访问网站的人。我正在跟踪数字数据，例如页面停留时间和其他数据。分类数据包括:位置(超过 1000 个唯一值)、Page_id(> 1000 个唯一值)、Author_id(超过 100 个唯一值)。最简单的解决方案是对所有内容进行单热编码，并将其放入具有可变序列长度的 LSTM 中，每个时间步对应不同的页面 View 。

上述DataFrame会生成7个训练样本，序列长度可变。例如，对于 user_id=2，我将有 2 个训练样本:

[ ROW_INDEX_4 ] and [ ROW_INDEX_4, ROW_INDEX_5 ]

设X为训练数据，我们来看第一个训练样本X[0]。

从上图中，我的分类特征是X[0][:, n:]。

在创建序列之前，我使用 pd.factorize() 将分类变量分解为 [0,1... number_of_cats-1]，因此 中的数据code>X[0][:, n:] 是其索引对应的数字。

我是否需要为每个分类特征分别创建一个嵌入？例如。 x_*n, x_*n+1, ..., x_*m 中的每一个的嵌入？

如果是这样，我如何将其放入 Keras 代码中？

model = Sequential()

model.add(Embedding(?, ?, input_length=variable)) # How do I feed the data into this embedding? Only the categorical inputs.

model.add(LSTM())
model.add(Dense())
model.add.Activation('sigmoid')
model.compile()

model.fit_generator() # fits the `X[i]` one by one of variable length sequences.

我的解决思路:

看起来像的东西:

我可以在每个分类特征 (m-n) 上训练一个 Word2Vec 模型来向量化任何给定的值。例如。伦敦将在 3 个维度上进行矢量化。假设我使用 3 维嵌入。然后我会把所有东西都放回 X 矩阵，它现在有 n + 3(n-m)，并使用 LSTM 模型来训练它？

我只是认为应该有更简单/更智能的方法。

Answer 1

正如您提到的，一种解决方案是对分类数据进行单热编码(或者甚至按原样使用它们，以基于索引的格式)并将它们与数值数据一起提供给 LSTM 层。当然，你也可以在这里有两个 LSTM 层，一个用于处理数值数据，另一个用于处理分类数据(以单热编码格式或基于索引的格式)，然后合并它们的输出。

另一种解决方案是为每个分类数据设置一个单独的嵌入层。每个嵌入层都可能有自己的嵌入维度(如上文所述，您可能有多个 LSTM 层来分别处理数值和分类特征):

num_cats = 3 # number of categorical features
n_steps = 100 # number of timesteps in each sample
n_numerical_feats = 10 # number of numerical features in each sample
cat_size = [1000, 500, 100] # number of categories in each categorical feature
cat_embd_dim = [50, 10, 100] # embedding dimension for each categorical feature

numerical_input = Input(shape=(n_steps, n_numerical_feats), name='numeric_input')
cat_inputs = []
for i in range(num_cats):
    cat_inputs.append(Input(shape=(n_steps,1), name='cat' + str(i+1) + '_input'))

cat_embedded = []
for i in range(num_cats):
    embed = TimeDistributed(Embedding(cat_size[i], cat_embd_dim[i]))(cat_inputs[i])
    cat_embedded.append(embed)

cat_merged = concatenate(cat_embedded)
cat_merged = Reshape((n_steps, -1))(cat_merged)
merged = concatenate([numerical_input, cat_merged])
lstm_out = LSTM(64)(merged)

model = Model([numerical_input] + cat_inputs, lstm_out)
model.summary()

这是模型摘要:

Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
cat1_input (InputLayer)         (None, 100, 1)       0                                            
__________________________________________________________________________________________________
cat2_input (InputLayer)         (None, 100, 1)       0                                            
__________________________________________________________________________________________________
cat3_input (InputLayer)         (None, 100, 1)       0                                            
__________________________________________________________________________________________________
time_distributed_1 (TimeDistrib (None, 100, 1, 50)   50000       cat1_input[0][0]                 
__________________________________________________________________________________________________
time_distributed_2 (TimeDistrib (None, 100, 1, 10)   5000        cat2_input[0][0]                 
__________________________________________________________________________________________________
time_distributed_3 (TimeDistrib (None, 100, 1, 100)  10000       cat3_input[0][0]                 
__________________________________________________________________________________________________
concatenate_1 (Concatenate)     (None, 100, 1, 160)  0           time_distributed_1[0][0]         
                                                                 time_distributed_2[0][0]         
                                                                 time_distributed_3[0][0]         
__________________________________________________________________________________________________
numeric_input (InputLayer)      (None, 100, 10)      0                                            
__________________________________________________________________________________________________
reshape_1 (Reshape)             (None, 100, 160)     0           concatenate_1[0][0]              
__________________________________________________________________________________________________
concatenate_2 (Concatenate)     (None, 100, 170)     0           numeric_input[0][0]              
                                                                 reshape_1[0][0]                  
__________________________________________________________________________________________________
lstm_1 (LSTM)                   (None, 64)           60160       concatenate_2[0][0]              
==================================================================================================
Total params: 125,160
Trainable params: 125,160
Non-trainable params: 0
__________________________________________________________________________________________________

然而，您可以尝试另一种解决方案:只为所有分类特征设置一个嵌入层。但它涉及一些预处理:您需要重新索引所有类别以使它们彼此不同。例如，第一个分类特征中的类别将从 1 编号到 size_first_cat，然后第二个分类特征中的类别将从 size_first_cat + 1 编号到 size_first_cat + size_second_cat 等等。然而，在这个解决方案中，所有分类特征都将具有相同的嵌入维度，因为我们只使用一个嵌入层。

---

更新:现在我想到了，您还可以在数据预处理阶段甚至在模型中 reshape 分类特征，以摆脱 TimeDistributed 层和Reshape 层(这也可能会提高训练速度):

numerical_input = Input(shape=(n_steps, n_numerical_feats), name='numeric_input')
cat_inputs = []
for i in range(num_cats):
    cat_inputs.append(Input(shape=(n_steps,), name='cat' + str(i+1) + '_input'))

cat_embedded = []
for i in range(num_cats):
    embed = Embedding(cat_size[i], cat_embd_dim[i])(cat_inputs[i])
    cat_embedded.append(embed)

cat_merged = concatenate(cat_embedded)
merged = concatenate([numerical_input, cat_merged])
lstm_out = LSTM(64)(merged)

model = Model([numerical_input] + cat_inputs, lstm_out)

至于拟合模型，您需要分别为每个输入层提供其对应的 numpy 数组，例如:

X_tr_numerical = X_train[:,:,:n_numerical_feats]

# extract categorical features: you can use a for loop to this as well.
# note that we reshape categorical features to make them consistent with the updated solution
X_tr_cat1 = X_train[:,:,cat1_idx].reshape(-1, n_steps) 
X_tr_cat2 = X_train[:,:,cat2_idx].reshape(-1, n_steps)
X_tr_cat3 = X_train[:,:,cat3_idx].reshape(-1, n_steps)

# don't forget to compile the model ...

# fit the model
model.fit([X_tr_numerical, X_tr_cat1, X_tr_cat2, X_tr_cat3], y_train, ...)

# or you can use input layer names instead
model.fit({'numeric_input': X_tr_numerical,
           'cat1_input': X_tr_cat1,
           'cat2_input': X_tr_cat2,
           'cat3_input': X_tr_cat3}, y_train, ...)

如果您想使用 fit_generator() 没有区别:

# if you are using a generator
def my_generator(...):

    # prep the data ...

    yield [batch_tr_numerical, batch_tr_cat1, batch_tr_cat2, batch_tr_cat3], batch_tr_y

    # or use the names
    yield {'numeric_input': batch_tr_numerical,
           'cat1_input': batch_tr_cat1,
           'cat2_input': batch_tr_cat2,
           'cat3_input': batch_tr_cat3}, batch_tr_y

model.fit_generator(my_generator(...), ...)

# or if you are subclassing Sequence class
class MySequnece(Sequence):
    def __init__(self, x_set, y_set, batch_size):
        # initialize the data

    def __getitem__(self, idx):
        # fetch data for the given batch index (i.e. idx)

        # same as the generator above but use `return` instead of `yield`

model.fit_generator(MySequence(...), ...)