thtrieu / qclass_dl Goto Github PK

After fix some place caused by different version of tf, I ran lstm model success but only get 84% Valid Acc, What do I missed?

Can you please give a script for testing the trained model for a new question(without a question type) from a console?

Why are you added the output of l2_loss(b) to l2_loss(self.w) ? It didnt make sense to me. Could you please explain what am I missing ?

l2_loss += tf.nn.l2_loss(self.W)
l2_loss += tf.nn.l2_loss(b)

Thanks.

Hi thtrieu:
Your repository is pretty cool and easy to read. But I have a few questions.

# Collect the appropriate last words into variable output (dimension = batch x embedding_size)
        output = outputs[0]
        one = tf.ones([1, hidden_unit], tf.float32)
        with tf.variable_scope("Output"):
            tf.get_variable_scope().reuse_variables()
            for i in range(1,len(outputs)):
                ind = self.real_len < (i+1)
                ind = tf.to_float(ind)
                ind = tf.expand_dims(ind, -1)
                mat = tf.matmul(ind, one)
                output = tf.add(tf.mul(output, mat),tf.mul(outputs[i], 1.0 - mat))

I am not sure why you do these step. Can you explain some detail about it please?

By the way, when I try to remove these lines, just use output = outputs[-1], I get terrible result. I guess output = outputs[-1] my be the output of <PAD/> word. So my result is terrible. Am I right?

Hello Sir, when i'm trying to run your program n Phyton 3, it gives me errors in this function.

def add_unknown_words(word_vecs, vocab, min_df=0, k=300):
"""
0.25 is chosen so the unknown vectors have (approximately) same variance as pre-trained ones
"""
count = 0
for word in vocab:
if word not in word_vecs and vocab[word] >= min_df:
word_vecs[word] = np.random.uniform(-0.25, 0.25, k)
else:
count += 1
return count

But when I'm trying to run on Python 3.6 it doesn't work and it makes the following errors. What should I change to run it on Python 3. Thanks

Traceback (most recent call last):
File "/Users/apple-PC/PycharmProjects/qclass_dl/train_cnn.py", line 56, in
count = data_helpers.add_unknown_words(trained_vecs, vocabulary)
File "/Users/apple-PC/PycharmProjects/qclass_dl/data_helpers.py", line 178, in add_unknown_words
word_vecs[word] = np.random.uniform(-0.25, 0.25, k)
TypeError: list indices must be integers or slices, not str