search

Conchylicultor / DeepQA

My tensorflow implementation of "A neural conversational model", a Deep learning based chatbot
Apache License 2.0
2.93k stars 1.17k forks source link

Why PreTrained model is not working ? #149

Open sushmas88 opened 7 years ago

sushmas88 commented 7 years ago

Here, in readme.md file you mentioned one link for pretrained model, we downloaded that file, and performed testing on that, but for my questions, bot didn't give correct answers, Even for "hi" message also its giving wrong result. what is the reason? is it a right pretrained model for generative bot? Can you suggest me how to workout this?

EMCP commented 7 years ago

what version of TF are you using? what command(s) did you use to invoke the interactive testing?
Also please include the exact output you did get..

my impression is the pre-trained bots sometimes do not get picked up if you mis-type something.. or place the file in the wrong folder. Seeing the output of your command can help verify it's properly utilizing the pretrained model.

arsi-shiva commented 7 years ago

the command that i am running and the output that i am getting are:

arsi$ python main.py --modelTag pretrainedv2 --test interactive Welcome to DeepQA v0.1 !

TensorFlow detected: v1.0.0 Loading dataset from /Users/arsi/git/data/samples/dataset-cornell-length10-filter1-vocabSize40000.pkl Loaded cornell: 24643 words, 159657 QA Model creation... W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations. Initialize variables... No previous model found, starting from clean directory: /Users/arsi/git/save/model-pretrainedv2 Testing: Launch interactive mode:

Welcome to the interactive mode, here you can ask to Deep Q&A the sentence you want. Don't have high expectation. Type 'exit' or just press ENTER to quit the program. Have fun. Q: hi A: Stooly negroes negroes negroes negroes negroes negroes oasis oasis oasis oasis intelligent

Q: hello A: Paso paso negroes negroes negroes negroes pronounced pronounced penicillin penicillin penicillin penicillin

Q: My name is David. What is my name ? A: 'was abnormal abnormal abnormal abnormal abnormal abnormal cloaking cloaking cloaking cloaking cloaking

arsi-shiva commented 7 years ago

Facing the similar issue, where while using the pre-trained model, the output is repetitive words, which seem irrelevant TF is 1.0.0

nltk (3.2.5) numpy (1.13.3) pip (9.0.1) protobuf (3.5.0) setuptools (36.8.0) six (1.11.0) tensorflow (1.0.0) textdata (1.7.3) tqdm (4.19.4) wheel (0.30.0)

Python 3.6.3

arsi-shiva commented 7 years ago

@EMCP any suggestions?

EMCP commented 7 years ago

make sure you aren't starting from scratch.. according to the output it looks like it didn't find the model

Initialize variables...
No previous model found, starting from clean directory: /Users/arsi/git/save/model-pretrainedv2

this is bad, and it means it's starting over I think..

wiwengweng commented 6 years ago

@EMCP Hi, so did you solve this problem?? Should I train my own model, and how to use the pre-trained model then? Thanks

EMCP commented 6 years ago

@wiwengweng I never had the problem.. but was trying to help others check their own setups.

I didn't focus on using the pre-trained models because I am using a newer version of Tensorflow than the models supported. Attempting to upgrade the pretrained models failed for me, and I just ignored that and moved onto other pursuits.

just make sure you're not starting from scratch by reading the console output

wiwengweng commented 6 years ago

So， obviously, I am from scratch... :-D I will get two type of work-around: 1) get Nicholas C.'s work and go on and on with my own businesshttps://mcastedu-my.sharepoint.com/personal/nicholas_cutajar_a100636_mcast_edu_mt/_layouts/15/guestaccess.aspx?folderid=077576c4cf9854642a968f67909380f45&authkey=AVt2JWMPkf2R_mWBpI1eAUY 2) traind a model to make everything work from scratch. But I now accept the first one, to set up demo. models in my own language will be trained later on. @EMCP Did you get any other suggestion?

EMCP commented 6 years ago

@wiwengweng I have no new things to add.. I just think it's worth sticking to the latest version of TF as possible when you're new...

if you get stuck in 1.0 , you're missing out on a lot of the new work coming out. I never got the translation tool to work on the pre-trained so I just moved on

wiwengweng commented 6 years ago

OK. Since the demo works for me, I just move on too. Thank you for the suggestion.