如何在 Tensorflow RNN 中构建嵌入层? [英] How to build an embedding layer in Tensorflow RNN?

问题描述

我正在构建一个 RNN LSTM 网络，以根据作者的年龄(二元分类 - 年轻/成人)对文本进行分类.

I'm building an RNN LSTM network to classify texts based on the writers' age (binary classification - young / adult).

似乎网络没有学习，突然开始过度拟合:

Seems like the network does not learn and suddenly starts overfitting:

红色:火车
蓝色:验证

一种可能是数据表示不够好.我只是按照它们的频率对独特的词进行排序并给它们索引.例如:

One possibility could be that the data representation is not good enough. I just sorted the unique words by their frequency and gave them indices. E.g.:

unknown -> 0
the     -> 1
a       -> 2
.       -> 3
to      -> 4

所以我试图用词嵌入替换它.我看到了几个例子，但我无法在我的代码中实现它.大多数示例如下所示:

So I'm trying to replace that with word embedding. I saw a couple of examples but I'm not able to implement it in my code. Most of the examples look like this:

embedding = tf.Variable(tf.random_uniform([vocab_size, hidden_size], -1, 1))
inputs = tf.nn.embedding_lookup(embedding, input_data)

这是否意味着我们正在构建一个学习嵌入的层?我认为应该下载一些 Word2Vec 或 Glove 并使用它们.

Does this mean we're building a layer that learns the embedding? I thought that one should download some Word2Vec or Glove and just use that.

无论如何，假设我想构建这个嵌入层...
如果我在代码中使用这两行，我会收到一个错误:

Anyway let's say I want to build this embedding layer...
If I use these 2 lines in my code I get an error:

类型错误:传递给参数索引"的值的数据类型 float32 不在允许值列表中:int32、int64

TypeError: Value passed to parameter 'indices' has DataType float32 not in list of allowed values: int32, int64

所以我想我必须将 input_data 类型更改为 int32.所以我这样做了(毕竟都是指数)，我明白了:

So I guess I have to change the input_data type to int32. So I do that (it's all indices after all), and I get this:

类型错误:输入必须是一个序列

TypeError: inputs must be a sequence

我尝试用一​​个列表包装 inputs(tf.contrib.rnn.static_rnn 的参数):[inputs] 如 这个答案，但这又产生了另一个错误:

I tried wrapping inputs (argument to tf.contrib.rnn.static_rnn) with a list: [inputs] as suggested in this answer, but that produced another error:

ValueError:输入大小(输入的维度 0)必须可以通过形状推断，但看到值无.

ValueError: Input size (dimension 0 of inputs) must be accessible via shape inference, but saw value None.

<小时>

更新:

在将张量 x 传递给 embedding_lookup 之前，我将其拆开.嵌入后我移动了拆垛.

I was unstacking the tensor x before passing it to embedding_lookup. I moved the unstacking after the embedding.

更新代码:

MIN_TOKENS = 10
MAX_TOKENS = 30
x = tf.placeholder("int32", [None, MAX_TOKENS, 1])
y = tf.placeholder("float", [None, N_CLASSES]) # 0.0 / 1.0
...
seqlen = tf.placeholder(tf.int32, [None]) #list of each sequence length*
embedding = tf.Variable(tf.random_uniform([VOCAB_SIZE, HIDDEN_SIZE], -1, 1))
inputs = tf.nn.embedding_lookup(embedding, x) #x is the text after converting to indices
inputs = tf.unstack(inputs, MAX_POST_LENGTH, 1)
outputs, states = tf.contrib.rnn.static_rnn(lstm_cell, inputs, dtype=tf.float32, sequence_length=seqlen) #---> Produces error

*seqlen:我对序列进行了零填充，因此它们都具有相同的列表大小，但由于实际大小不同，我准备了一个描述没有填充的长度的列表.

新错误:

ValueError: 层 basic_lstm_cell_1 的输入 0 与图层:预期 ndim=2，发现 ndim=3.收到完整形状:[无，1, 64]

ValueError: Input 0 of layer basic_lstm_cell_1 is incompatible with the layer: expected ndim=2, found ndim=3. Full shape received: [None, 1, 64]

64 是每个隐藏层的大小.

很明显，我的维度有问题......嵌入后如何使输入适合网络?

It's obvious that I have a problem with the dimensions... How can I make the inputs fit the network after embedding?

推荐答案

来自 tf.nn.static_rnn ，我们可以看到 inputs 参数是:

From the tf.nn.static_rnn , we can see the inputs arguments to be:

一个长度为 T 的输入列表，每个输入一个形状为 [batch_size, input_size] 的张量

A length T list of inputs, each a Tensor of shape [batch_size, input_size]

所以你的代码应该是这样的:

So your code should be something like:

x = tf.placeholder("int32", [None, MAX_TOKENS])
...
inputs = tf.unstack(inputs, axis=1)

这篇关于如何在 Tensorflow RNN 中构建嵌入层?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！