search

ugroot / GROOT

GROOT: Welcome to my source code, I'm your personal assistant
Mozilla Public License 2.0
27 stars 28 forks source link

We need to make a model and train it so that we can reply in more stunning way #4

Open omi10859 opened 6 years ago

omi10859 commented 6 years ago

This issue is open for discussion

ConformalTomb commented 6 years ago

So I was thinking to use a pre-existing dataframe of questions and answers for speech like assistance and Siri offers. After we find that data (which I don't think will be easy) we can implement that to predict better answers.

omi10859 commented 6 years ago

ok, so someone has already extracted reddit (nearly 40 million lines of Q/A) for the same we can use that.

omi10859 commented 6 years ago

and there is no pre-existing dataframe for Q/A. Siri and all of them use Deep learning to generate an answer using data of similar type given to them... @ConformalTomb

ConformalTomb commented 6 years ago

Well then let's go the deep learning way! Earlier I was thinking of the decision tree model for starters.

rohit6205 commented 5 years ago

Are you done with this issue or still working ??

omi10859 commented 5 years ago

Nupe we need help @rohit6205

d3sm0 commented 5 years ago

At what point are you? Did you find a model to implement /use? Is there a branch of this?

rohit6205 commented 5 years ago

I was working on this for my project and I think, this kind of chatbot will work for us now. because dataset is accessible and model creation is easy. But to implement this we need more computational power (GPU).

omi10859 commented 5 years ago

More computational power for training model? We can do that on servers

sarthak247 commented 5 years ago

I can work on this but I'm having my sessionals right now. Also, for the computational part, we can use Google Colab. It provides 12 hours for running a kernel. And Kaggle provides 6 hours to run a kernel.

arnabsinha99 commented 5 years ago

I want to contribute to this issue along with anyone who is doing it? Is it possible?

omi10859 commented 5 years ago

yeah.. join our slack channel for more!

sarthak247 commented 5 years ago

I want to contribute to this issue along with anyone who is doing it? Is it possible?

Well. I'm doing this but you won't be able to contribute to it as of now. You'll have to wait till 16 March for me to commit the chatbot so that you can later contribute to it by making changes to parameters etc

The reason for delay is that I'm having my sessionals now and also training the model would take 2-3 days due to high computational needs. I've already collected the data, filtered it. Now only the training part is left. I'll start that on 14 March and commit the chatbot by 16 or at max 17 March (if it is trained by that time)

aish-where-ya commented 5 years ago

We could use Google's Dialogflow to train the conversational model.

sarthak247 commented 5 years ago

We could use Google's Dialogflow to train the conversational model.