我正在尝试多类分类，这里是我的训练输入和输出的详细信息。我看过其他帖子/答案，但无法弄清楚。感谢您的帮助。
**vectorized_data['train_sents'] #This is how the train data looks like
array([[[ 1, 1, 1, ..., 1, 1, 1],
[ 0, 0, 0, ..., 0, 0, 0],
[5818, 7142, 2467, ..., 7927, 7927, 7927],
[ 10, 19, 6, ..., 22, 22, 22],
[ 6, 6, 6, ..., 8, 8, 8],
[ 0, 0, 0, ..., 7927, 7927, 7927]],
[[1108, 1108, 1108, ..., 1108, 1108, 1108],
[ 0, 0, 0, ..., 0, 0, 0],
[2244, 1362, 904, ..., 7927, 7927, 7927],
[ 0, 0, 13, ..., 22, 22, 22],
[ 3, 9, 6, ..., 8, 8, 8],
[ 0, 0, 0, ..., 7927, 7927, 7927]]])
vectorized_data['train_sents'].shape
(2522, 6, 67)
vectorized_data['train_lab'] # Labels data
array([[46, 46, 7, ..., 25, 25, 25],
[46, 46, 46, ..., 25, 25, 25],
[46, 11, 3, ..., 25, 25, 25],
...,
[46, 46, 46, ..., 25, 25, 25],
[ 7, 31, 31, ..., 25, 25, 25],
[ 7, 31, 46, ..., 25, 25, 25]])
vectorized_data['train_lab'].shape
(2522, 67)
#Define Batch size
train_batch_size = 32
valid_batch_size = 64
test_batch_size = 64
max_seq_len = 124 #Define max_length of 124 for padding sentences and labels
EMBEDDING_DIM = 300
epochs = 5
hidden = 256 #no of neurons
lr = .0001 # learning rate
batched_train_dataset
<BatchDataset shapes: ((32, 6, 67), (32, 67)), types: (tf.int64, tf.int64)>
#Define RNN Simple model
class RNN_Simple(tf.keras.Model):
def __init__(self, max_seq_len, embed_input_dim, embed_output_dim, num_labels, weights):
super(RNN_Simple, self).__init__()
self.embedding = layers.Embedding(input_dim=embed_input_dim, output_dim=embed_output_dim,
weights=weights, input_length=max_seq_len, trainable=False, mask_zero=True)
self.rnnsimple = layers.SimpleRNN(hidden, return_sequences=True) #Define hidden layer with 256 units
self.dense = layers.Dense(num_labels, activation = 'softmax')
def call(self, inputs):
x = self.embedding(inputs)
x = self.rnnsimple(x)
logits = self.dense(x)
return logits
# Run RNN Simple model
model = RNN_Simple(max_seq_len=max_seq_len,embed_input_dim=len(vocab), embed_output_dim=EMBEDDING_DIM, weights=[embed_weights], num_labels=len(indicies))
optimizer = tf.keras.optimizers.Adam(learning_rate=lr) #use Adam Optimizer with learning rate = .0001
sccel = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) #sparse categorical cross-entropy loss
# Train the RNN Simpl model. Calculate Loss.
train_loss_m = tf.keras.metrics.Mean('training_loss', dtype=tf.float32)
valid_loss_m = tf.keras.metrics.Mean('valid_loss', dtype=tf.float32)
def train_func(sent_batch, labels_batch):
'''
This function applies gradients and calculates loss for training dataset
'''
with tf.GradientTape() as tape:
logits = model(sent_batch)
loss = sccel(labels_batch, logits)
grads = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(list(zip(grads, model.trainable_variables)))
return loss, logits**
此功能在 >>>>>> logits = model(sent_batch)
上失败 def train_func(sent_batch, labels_batch):
with tf.GradientTape() as tape:
logits = model(sent_batch)
错误是： ValueError: 层 simple_rnn_16 的输入 0 与层不兼容：预期 ndim=3，发现 ndim=4。收到完整形状：(32, 6, 67, 300)