Bi-LSTM with Keras : dimensions must be equal but are 7 and 300-CodePudding

I am creating for the first time a bilstm with keras but I am having difficulties. So that you understand, here are the steps I have done:

I created an embedding matrix with Glove for my x ;

def create_embeddings(fichier,dictionnaire,dictionnaire_tokens):
    with open(fichier) as file:
        line = file.readline()
        max_words = max(dictionnaire_tokens.values()) 1 #1032
        max_size_dimensions = 300
        emb_matrix = np.zeros((max_words,max_size_dimensions))
        for item,count in dictionnaire_tokens.items():
            try:
                vecteur = dictionnaire[item]
            except:
                pass
            if vecteur is not None:
                emb_matrix[count]= vecteur
        return emb_matrix

I did some one hot encoding with my y's;

def one_hot_encoding(file):
    with open(file) as file:
        line = file.readline()
        liste = []
        while line:
            tag = line.split(" ")[1]
            tag = [tag]
            line = file.readline()
            liste.append(tag)
    one_hot = MultiLabelBinarizer()
    array = one_hot.fit_transform(liste)
    return array

I compiled my model with keras

from tensorflow.keras.layers import Bidirectional
model = Sequential()

embedding_layer = Embedding(input_dim=1031   1,
                                    output_dim=300,
                                    weights=[embedding_matrix],
                                    trainable=False)
model.add(embedding_layer)

bilstm_layer = Bidirectional(LSTM(units=300, return_sequences=True))
model.add(bilstm_layer)
model.add(Dense(300, activation="relu"))
#crf_layer = CRF(units=len(self.tags), sparse_target=True)
#model.add(crf_layer)
model.compile(optimizer="adam", loss='binary_crossentropy', metrics='acc')
model.summary()

Input of my embedding layer (embedding matrix) : 

[[ 0.        0.        0.       ...  0.        0.        0.      ]
 [ 0.        0.        0.       ...  0.        0.        0.      ]
 [ 0.        0.        0.       ...  0.        0.        0.      ]
 ...
 [-0.068577 -0.71314   0.3898   ... -0.077923 -1.0469    0.56874 ]
 [ 0.32461   0.50463   0.72544  ...  0.17634  -0.28961   0.29007 ]
 [-0.33771  -0.24912  -0.032685 ... -0.033254 -0.45513  -0.13319 ]]

I train my model. However when I want to train it, I get the following message: ValueError: Dimensions must be equal, but are 7 and 300 for '{{node binary_crossentropy/mul}} = Mul[T=DT_FLOAT](binary_crossentropy/Cast, binary_crossentropy/Log)' with input shapes: [?,7], [?,300,300].

My embedding matrix was made with glove 300d so it has 300 dimensions. While my labels, I have only 7 labels. So I have to make my x and y have the same dimensions but how? Thank you!!!

CodePudding user response：

keras.backend.clear_session()
from tensorflow.keras.layers import Bidirectional
model = Sequential()

_input = keras.layers.Input(shape=(300,1))
model.add(_input)
bilstm_layer = Bidirectional(LSTM(units=300, return_sequences=False))
model.add(bilstm_layer)
model.add(Dense(7, activation="relu")) #here 7 is the number of classes you have and None is the batch_size
#crf_layer = CRF(units=len(self.tags), sparse_target=True)
#model.add(crf_layer)
model.compile(optimizer="adam", loss='binary_crossentropy', metrics='acc')
model.summary()