原因：损失：nan acc:0.0000e+00 val_损失：nan val_acc:0.0000e+00 | TensorFlow 2.0 | Python

Question

我正在使用一个LSTM模型，根据迈尔斯-布里格斯测试预测一种人格类型

• 我创建了一个数据集（295,2（不包括标题）），其中包含个性类型标签及其各自的描述

csv文件：Dataset MBTI |Github

• 数据集分为各自的矩阵：

  80%(train_data, train_labels) | 20%(validation_data, validation_labels).

• 此外，对数据集进行了预处理：标记化、stopwords和填充

但是，在培训时，如下所示：

Train on 236 samples, validate on 59 samples
Epoch 1/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 2/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 3/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 4/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 5/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 6/100
236/236 - 2s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 7/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 8/100
236/236 - 2s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 9/100
236/236 - 2s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00
Epoch 10/100
236/236 - 1s - loss: nan - acc: 0.0000e+00 - val_loss: nan - val_acc: 0.0000e+00

LSTM模型如下所示：

model = keras.Sequential()
model.add(keras.layers.Embedding(600, 295))
model.add(keras.layers.Bidirectional(keras.layers.LSTM(590))) model.add(keras.layers.GlobalAveragePooling1D())
model.add(keras.layers.Dense(16, activation = 'relu'))
model.add(keras.layers.Dense(16, activation = 'sigmoid')) # Not sure if using sigmoid | softmax

model.summary()
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
embedding (Embedding)        (None, None, 295)         147500    
_________________________________________________________________
bidirectional (Bidirectional (None, 64)                83968     
_________________________________________________________________
dense (Dense)                (None, 16)                1040      
_________________________________________________________________
dense_1 (Dense)              (None, 16)                272       
=================================================================
Total params: 232,780
Trainable params: 232,780
Non-trainable params: 0

预期输出：

• 我希望在训练时，投出一杯“95%”的准确度，以及与其余参数（loss、val_loss、val_acc）相对应的正常值
• 如果模型不正确，或者​​不是你应该输入的，让我知道，像任何其他建议，以更好地理解问题和促进答案