How predict all responses at one time? about chatbot-retrieval HOT 6 CLOSED

Thank you, im my fork I have implemented a second option, similar to the test mode

https://github.com/Sangheli/chatbot-retrieval

from chatbot-retrieval.

Two options:

• You could try using the X and y arguments you mention above instead of the input_fn. Then you can make X an iterator of examples instead of just a single one.
• You could modify the get_features function to use a Queue and produce multiple examples instead of a single one

from chatbot-retrieval.

@e-budur This is the case for Sangheli's version of prediction. It always takes the first response as the best one, but when I give the potential_responses one by one to the estimator as it is in Denny's original prediction code, it works correctly but it is as slow as the prediction takes 4 minutes for 500 potential responses!!!

from chatbot-retrieval.

@Samurais thanks for posting your branch! I had the same question, very helpful.

from chatbot-retrieval.

@Sangheli In your prediction code (your udc_predict.py), it is true that you are sending all the potential responses to the estimator at once but the answer is always the first response in the potential responses simply because the estimator ignore other elements in the potential responses list. In fact, although you give the features extracted from whole potential responses into the estimator but based on my observations it always gives the first element in the potential_responses as an answer. However, if you put the estimator in the loop exactly like what it is in Denny's code, you will get the correct response but unfortunately, it takes very long time to get the response for 500 potential responses....

from chatbot-retrieval.

from chatbot-retrieval.