dongjun-lee / text-classification-models-tf Goto Github PK

data_utils.py function build_char_dataset

some seem not very reasonable:

when model = "char_cnn":  
    char_dict["<pad>"] = char_dict["<unk>"] = char_dict["a"] = 0 after onehot in char_cnn.py

when model = "vd_cnn": 
    char_dict["<pad>"] = char_dict["a"] = 0, 
    char_dict["<unk>"] = char_dict["b"] = 1

Have you test the SST-2 dataset ?
I have run word_cnn model on SST-2 dataset . Only get 0.38 accuracy. But the accuray is 0.45 in paper.
How to change the model to improve the accuracy? Is something wrong?
The word_rnn model can get 0.45 accuracy, why diff is so large?

Tensorflow 2.x
Python 3.7.3

Traceback (most recent call last):
File "train3.py", line 74, in
model = WordCNN(vocabulary_size, WORD_MAX_LEN, NUM_CLASS)
File "/home/migueltuxd/bucket4testingtpuss/TPU/text-classification-models-tf-master/cnn_models/word_cnn.py", line 54, in init
self.optimizer = tf.train.AdamOptimizer(self.learning_rate).minimize(self.loss, global_step=self.global_step)
File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/training/optimizer.py", line 413, in minimize
name=name)
File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/training/optimizer.py", line 564, in apply_gradients
raise RuntimeError("Use _distributed_apply() instead of "
RuntimeError: Use _distributed_apply() instead of apply_gradients() in a cross-replica context.

Shared too on stackoverflow:https://stackoverflow.com/questions/61704387/wordcnn-trouble-with-distributed-apply-and-apply-gradients

How can I get prediction results in test.py?
The implementation now only has accuracy, so it's hard to see precision, recall, f1, or the prediction itself.

每次卷积输入的都是input，第二次循环也是conv变量，这样第一次循环不是就用不上了？

find a error when train the att_rnn model

Machine: 4 * Tesla P100-PCIE-16GB, memory: 256G

ResourceExhaustedError (see above for traceback): OOM when allocating tensor of shape [563354,256] and type float [[Node: embeddings/Adam/Initializer/zeros = Const[dtype=DT_FLOAT, value=Tensor<type: float shape: [563354,256] values: [0 0 0]...>, _device="/job:localhost/replica:0/task:0/device:GPU:0"]()]]

It crashes in Tensorflow 1.4 with the error below:

Traceback (most recent call last):
File "train.py", line 47, in
model = VDCNN(alphabet_size, CHAR_MAX_LEN, NUM_CLASS)
File "text-classification-models-tf/cnn_models/vd_cnn.py", line 70, in init
tf.nn.softmax_cross_entropy_with_logits_v2(logits=self.logits, labels=y_one_hot))
AttributeError: module 'tensorflow.python.ops.nn' has no attribute 'softmax_cross_entropy_with_logits_v2'

how to predict for one sentence?