search

shahsohil / DCC

This repository contains the source code and data for reproducing results of Deep Continuous Clustering paper
MIT License
208 stars 53 forks source link

Hyperparams and resulting numbers #33

Open meitarronen opened 2 years ago

meitarronen commented 2 years ago

Hi, thank you so much for your contribution and sharing your project! This is a great paper :) I would highly appreciate your help with running the training process as I am not sure how to run certain parts of it in order to reproduce your results from the paper.

  1. How should we run the edgeConstruction script? Which hyperparams should we choose?
  2. Do you have available configurations for more datasets?
  3. Using the following commands I got fairly good numbers on MNIST, but lower than the reported ones. Can you perhaps guide me how to get closer to the reported numbers?

Script lines: python pretraining.py --data mnist --id 1 --niter 50000 --lr 10 --step 20000 python extract_feature.py --data mnist --net checkpoint_4.pth.tar --features pretrained python edgeConstruction.py --dataset mnist --format mat --samples 70000 --prep 'minmax' --k 10 --algo 'mknn' python copyGraph.py --data mnist --graph pretrained.mat --features pretrained.pkl --out pretrained python DCC.py --data mnist --net checkpoint_4.pth.tar --id 1 The results I got: ARI: 0.830861826385 AMI: 0.7969629161498257 NMI: 0.8647221174121507 ACC: 0.8187 K: 173

Thank you so much in advance!!