parya-j / team-rope

This repository contains materials of final project for data science practicum offered at University of Georgia by Dr. Quinn Spring 2018
MIT License
0 stars 1 forks source link

Merge models #6

Open omid-s opened 6 years ago

omid-s commented 6 years ago

Models need to be merged in a reasonable state, we have the following options :

what's your thoughts ?

parya-j commented 6 years ago

How about running each model separately and use their outputs? I think the first approach is doable ..

raun1 commented 6 years ago

do we have the clustering and the pretrained models ready?

raun1 commented 6 years ago

We can train a final network on top of it

omid-s commented 6 years ago

kmeans is ready

raun1 commented 6 years ago

ok can u tell me the name of the file i will merge the dense net + k means and run a neural network on top of it :>

omid-s commented 6 years ago

kmeans http://cobweb.cs.uga.edu/~omid/models/kmeans_100_Classes.pkl

raun1 commented 6 years ago

@parya-j do u have the outputs for the pre trained network ready?

omid-s commented 6 years ago

assumes 64*64 input :) to load the model see : http://scikit-learn.org/stable/modules/model_persistence.html it uses joblib libjob what ever :D

parya-j commented 6 years ago

What type of output you expect? Now it gives you probability of classes of imagenet

omid-s commented 6 years ago

I think we should stick with probabilities, it will give us mupltiple objects in a scene

raun1 commented 6 years ago

@omid-s
We will do exactly what parya suggested ;) Clustering is unsupervised so no need to load the model. Just need the csv or some file for like

I already have them for dense nets.

@parya-j yes exactly basically the class prediction for each class. Use argmax to get the class (not probability) i.e. the class will be the arg max on softmax layer. Do this for both training and testing :)

raun1 commented 6 years ago

@omid-s for the testing set we should use the similar clusters tho :/ i guess we have to compute the shortest distance from the training centers and assign them. Let me check the source on that.

omid-s commented 6 years ago

I'll do that for parya's model if you do a argmax what you'll get is the one with the highest posibility! this will give us only one object in the photo ( most probably the person! ) if we stick with the probabs we can have a glimps of all the objects in the scene, like person, sea, tree, mountain! is far better than just person

raun1 commented 6 years ago

u have to retrain the top layers of the pretrained model tho

omid-s commented 6 years ago

let;s talk about it tomorrow, I do not follow why is that but passing out hate 8 am class :((

raun1 commented 6 years ago

Yea you have a point :) lets keep it all the things just give me the probability map of the softmax. my bad dont do argmax