chiphuyen/stanford-tensorflow-tutorials

i want to make feature to add custom training data to existing one

MartinAbilev opened this issue · 0 comments

Say i want model to get train on same voc and train made from cornel plus to train on my onv conversations from bot. maybe there is some recommendations how to do this in the best way ??

my current idea is to save similar file structure in given folder set in config.py. then on load just append
to existing train "cornel" ones. use same voc. and of course swap enc dec data to make my ansvers to bot "correct" ones. training data i leave as it is.

for me this feature is the essential to make experiments with "changing" model. i done something similar with "DeepQA" because there is just one vocab and litla less mess with files. But after some experimentation i found "original" Chips version is better ... and this separate enc dec actualy is the benefit. because word "me" when i say it to bot is something different when bot says it to me. i bet it solve reflection problem when bot starts to think he is me, and answers to what is your name "my name is martin" because i answers to him in such way.... if i get "fixed" my home pc i plan to get things up this weekend.