ajaech / query_completion Goto Github PK

I have trained model on my dataset but at the time of iferencing I'm getting value error.

ValueError: The last dimension of the inputs to a Dense layer should be defined. Found None. Full input shape received: <unknown>

how to inference .Do you give me your test data。 you can send me email [email protected]。thanks

in beam.py file

"class BeamItem " need to amend follows：

class BeamItem(object):
    """This is a node in the beam search tree.

    Each node holds four things: a log probability, a list of previous words, and
    the two hidden state vectors.
    """

    def __init__(self, prev_word, prev_hidden, log_prob=0.0):
        self.log_probs = log_prob
        if type(prev_word) == list:
            self.words = prev_word
        else:
            self.words = [prev_word]
        self.prev_hidden = prev_hidden

    **def __le__(self, other):
        return self.log_probs <= other.log_probs

    def __lt__(self, other):
        return self.log_probs < other.log_probs

    def __ge__(self, other):
        return self.log_probs >= other.log_probs

    def __gt__(self, other):
        return self.log_probs > other.log_probs**

    # 重定义等于比较
    def __eq__(self, other):
        return self.log_probs == other.log_probs

    def __str__(self):
        return "beam {0:.3f}: ".format(self.log_probs) + ''.join(self.words)

because PriorityQueue has Priority feature 。so ，it need add compare function。

in mode.py file

line 130 、 132 。

130: self.per_sentence_loss = tf.div(tf.reduce_sum(self.per_word_loss, 1),
tf.reduce_sum(_mask, 1))
132： self.per_sentence_loss = tf.reduce_sum(self.per_word_loss, 1)

here
why same code definition and it not used in after code

thanks

Can you guide me run your code? I don't know where to locate data and what expdir is.

Thanks for your sharing. I want to run your code.
Can you describe how to prepare data before running trainer.py (to reproduce your results)?

(1) Where can I download original data?
http://www.cim.mcgill.ca/~dudek/206/Logs/AOL-user-ct-collection/
Is this correct?

(2) Then, how to construct data.tsv, valdata.tsv, `testdata.tsv'?

Or can you provide codes for the split?