DCA - Official Python implementation of Delaunay Component Analysis algorithm

Related tags

Deep LearningDCA
Overview

Delaunay Component Analysis (DCA)

Official Python implementation of the Delaunay Component Analysis (DCA) algorithm presented in the paper Delaunay Component Analysis for Evaluation of Data Representations. If you use this code in your work, please cite it as follows:

Citation

@inproceedings{
    poklukar2022delaunay,
    title={Delaunay Component Analysis for Evaluation of Data Representations},
    author={Petra Poklukar and Vladislav Polianskii and Anastasiia Varava and Florian T. Pokorny and Danica Kragic Jensfelt},
    booktitle={International Conference on Learning Representations},
    year={2022},
    url={https://openreview.net/forum?id=HTVch9AMPa}
}

Getting started

Setup

Install the requirements with poetry:

poetry install
chmod +x dca/approximate_Delaunay_graph

Note: Delaunay graph building algorithm requires access to a GPU.

First example

  1. Run a 2D example that saves the intermediate files:
poetry run python examples/first_example.py 
  1. Check out the results saved output/first_example which will have the following structure:
experiments/first_example/
  /precomputed
    - clusterer.pkl               # HDBSCAN clusterer object
    - input_array.npy             # array of R and E points
    - input_array_comp_labels.npy # array of component labels corresponding to R and E points
    - unfiltered_edges.npy        # array of unfiltered approximated Delaunay edges
    - unfiltered_edges_len.npy    # array of unfiltered approximated Delaunay edge lengths
  /template_id1
    - output.json                 # dca scores 
    /DCA
        - components_stats.pkl    # Local evaluation scores
        - network_stats.pkl       # Global evaluation scores
    /visualization
        - graph visualizations
    /logs
        - version0_elapsed_time.log      # empirical runtime 
        - version0_input.json            # specific input parameters
        - version0_output_formatted.log  # all evaluation scores in a pretty format
        - version0_experiment_info.log   # console logs
        - # output files from qDCA
        - # any additional logs that should not be shared across experiment_ids in precomputed folder

Note: you can modify the experiment structure by definining what is shared across several experiments, e.g., what goes in the output/first_example/precomputed folder. For examples, see CL_ablation_study.py.

  1. In output/first_example/template_id1/visualization folder you should see an image of the approximated Delaunay graph and the distilled Delaunay graph like the ones below:

first_example

  1. In output/first_example/template_id1/logs/version0_output_formatted.log you should see the following output:
[mm/dd/yyyy hh:mm:ss] :: num_R: 20                            # total number of R points
[mm/dd/yyyy hh:mm:ss] :: num_E: 20                            # total number of E points
[mm/dd/yyyy hh:mm:ss] :: precision: 0.95                      
[mm/dd/yyyy hh:mm:ss] :: recall: 0.4
[mm/dd/yyyy hh:mm:ss] :: network_consistency: 1.0
[mm/dd/yyyy hh:mm:ss] :: network_quality: 0.2
[mm/dd/yyyy hh:mm:ss] :: first_trivial_component_idx: 2       # idx of the first outlier
[mm/dd/yyyy hh:mm:ss] :: num_R_points_in_fundcomp: 8          # number of vertices in F^R
[mm/dd/yyyy hh:mm:ss] :: num_E_points_in_fundcomp: 19         # number of vertices in F^E
[mm/dd/yyyy hh:mm:ss] :: num_RE_edges: 19                     # number of heterogeneous edges in G_DD
[mm/dd/yyyy hh:mm:ss] :: num_total_edges: 95                  # number of all edges in G_DD
[mm/dd/yyyy hh:mm:ss] :: num_R_outliers: 0                    
[mm/dd/yyyy hh:mm:ss] :: num_E_outliers: 1
[mm/dd/yyyy hh:mm:ss] :: num_fundcomp: 1                      # number of fundamental components |F|
[mm/dd/yyyy hh:mm:ss] :: num_comp: 3                          # number of all connected components
[mm/dd/yyyy hh:mm:ss] :: num_outliercomp: 1                   # number of trivial components
# Local scores for each component G_i: consistency and quality (Def 3.2) as well as number of R and E points contained in it
[mm/dd/yyyy hh:mm:ss] :: c(G0): 0.59, q(G0): 0.27, |G0^R|_v: 8   , |G0^E|_v: 19  , |G0|_v: 27  
[mm/dd/yyyy hh:mm:ss] :: c(G1): 0.00, q(G1): 0.00, |G1^R|_v: 12  , |G1^E|_v: 0   , |G1|_v: 12  
[mm/dd/yyyy hh:mm:ss] :: c(G2): 0.00, q(G2): 0.00, |G2^R|_v: 0   , |G2^E|_v: 1   , |G2|_v: 1   
  1. If you are only interested in the output DCA scores, the cleanup function will remove all of the intermediate files for you. Test it on this 2D example by running
poetry run python examples/first_example.py --cleanup 1

Note: to run q-DCA it is required to keep the intermediate files. This is because the distilled Delaunay graph is needed to calculate edges to the query points.

Run DCA on your own representations

Minimum example requires you to define the input parameters as in the code below. See dca/schemes.py for the optional arguments of the input configs.

# Generate input parameters
data_config = REData(R=R, E=E)
experiment_config = ExperimentDirs(
    experiment_dir=experiment_path,
    experiment_id=experiment_id,
)
graph_config = DelaunayGraphParams()
hdbscan_config = HDBSCANParams()
geomCA_config = GeomCAParams()

# Initialize loggers
exp_loggers = DCALoggers(experiment_config.logs_dir)

# Run DCA
dca = DCA(
    experiment_config,
    graph_config,
    hdbscan_config,
    geomCA_config,
    loggers=exp_loggers,
)
dca_scores = dca.fit(data_config)
dca.cleanup()  # Optional cleanup

Reproduce experiments in the paper

Datasets

We used and adjusted datasets used in our eariler work GeomCA. Therefore, we only provide the representations used in the contrastive learning experiment and q-DCA stylegan experiment, which you can download on this link and save them in representations/contrastive_learning and representations/stylegan folders, respectively. For VGG16, we provide the code (see VGG16_utils.py) we used on the splits constructed in GeomCA. For StyleGAN mode truncation experiment, we refer the user either to the splits we provided in GeomCA or to the code provided by Kynkäänniemi et. al.

Section 4.1: Contrastive Learning

Reproduce Varying component density experiment:

poetry run python experiments/contrastive_learning/CL_varying_component_density.py --n-iterations 10 --perc-to-discard 0.5 --cleanup 1

Reproduce Cluster assignment experiment, for example, using query set Q2 and considering flexible assignment procedure:

poetry run python experiments/contrastive_learning/CL_qDCA.py Df query_Df_holdout_c7_to_c11 --run-dca 1 --run-qdca 1 --several-assignments 1 --cleanup 1

Reproduce Mode truncation experiment in Appendix B.1:

poetry run python experiments/contrastive_learning/CL_mode_truncation.py --cleanup 1

Reproduce Ablation study experiments in Appendix B.1:

poetry run python experiments/contrastive_learning/CL_ablation_study.py cl-ablation-delaunay-edge-approximation --cleanup 1
poetry run python experiments/contrastive_learning/CL_ablation_study.py cl-ablation-delaunay-edge-filtering --cleanup 1
poetry run python experiments/contrastive_learning/CL_ablation_study.py cl-ablation-hdbscan --cleanup 1

Section 4.2: StyleGAN

Reproduce Mode truncation experiment, for example, on truncation 0.5 and 5000 representations provided by Poklukar et. al in GeomCA:

poetry run python experiments/stylegan/StyleGAN_mode_truncation.py 0.5 --num-samples "5000" --cleanup 1

Reproduce Quality of individual generated images experiment using qDCA, for example, on truncation 0.5 --cleanup 1

poetry run python experiments/stylegan/StyleGAN_qDCA.py --run-dca 1 --run-qdca 1 --cleanup 1

Section 4.3: VGG16

Reproduce Class separability experiment, for example, on version 1 containing classes of dogs and kitchen utils

poetry run python experiments/vgg16/VGG16_class_separability.py --version-id 1 --cleanup 1 

Reproduce Amending labelling inconsistencies experiment using qDCA, for example, on version 1 containing classes of dogs and kitchen utils

poetry run python experiments/vgg16/VGG16_qDCA.py --version-id 1 --run-dca 1 --run-qdca 1 --cleanup 1
Owner
Petra Poklukar
Petra Poklukar
Unimodal Face Classification with Multimodal Training

Unimodal Face Classification with Multimodal Training This is a PyTorch implementation of the following paper: Unimodal Face Classification with Multi

Wenbin Teng 3 Jul 06, 2022
PyTorch code for our paper "Gated Multiple Feedback Network for Image Super-Resolution" (BMVC2019)

Gated Multiple Feedback Network for Image Super-Resolution This repository contains the PyTorch implementation for the proposed GMFN [arXiv]. The fram

Qilei Li 66 Nov 03, 2022
Unofficial implementation of Pix2SEQ

Unofficial-Pix2seq: A Language Modeling Framework for Object Detection Unofficial implementation of Pix2SEQ. Please use this code with causion. Many i

159 Dec 12, 2022
Source code for Task-Aware Variational Adversarial Active Learning

Contrastive Coding for Active Learning under Class Distribution Mismatch Official PyTorch implementation of ["Contrastive Coding for Active Learning u

27 Nov 23, 2022
Official code for the paper "Self-Supervised Prototypical Transfer Learning for Few-Shot Classification"

Self-Supervised Prototypical Transfer Learning for Few-Shot Classification This repository contains the reference source code and pre-trained models (

EPFL INDY 44 Nov 04, 2022
Technical Analysis Indicators - Pandas TA is an easy to use Python 3 Pandas Extension with 130+ Indicators

Pandas TA - A Technical Analysis Library in Python 3 Pandas Technical Analysis (Pandas TA) is an easy to use library that leverages the Pandas package

Kevin Johnson 3.2k Jan 09, 2023
RTSeg: Real-time Semantic Segmentation Comparative Study

Real-time Semantic Segmentation Comparative Study The repository contains the official TensorFlow code used in our papers: RTSEG: REAL-TIME SEMANTIC S

Mennatullah Siam 592 Nov 18, 2022
Sequence-tagging using deep learning

Classification using Deep Learning Requirements PyTorch version = 1.9.1+cu111 Python version = 3.8.10 PyTorch-Lightning version = 1.4.9 Huggingface

Vineet Kumar 2 Dec 20, 2022
Official code for paper "ISNet: Costless and Implicit Image Segmentation for Deep Classifiers, with Application in COVID-19 Detection"

Official code for paper "ISNet: Costless and Implicit Image Segmentation for Deep Classifiers, with Application in COVID-19 Detection". LRPDenseNet.py

Pedro Ricardo Ariel Salvador Bassi 2 Sep 21, 2022
LaBERT - A length-controllable and non-autoregressive image captioning model.

Length-Controllable Image Captioning (ECCV2020) This repo provides the implemetation of the paper Length-Controllable Image Captioning. Install conda

bearcatt 53 Nov 13, 2022
In the case of your data having only 1 channel while want to use timm models

timm_custom Description In the case of your data having only 1 channel while want to use timm models (with or without pretrained weights), run the fol

2 Nov 26, 2021
face_recognization (FaceNet) + TFHE (HNP) + hand_face_detection (Mediapipe)

SuperControlSystem Face_Recognization (FaceNet) 面部识别 (FaceNet) Fully Homomorphic Encryption over the Torus (HNP) 环面全同态加密 (TFHE) Hand_Face_Detection (M

liziyu0104 2 Dec 30, 2021
(CVPR2021) ClassSR: A General Framework to Accelerate Super-Resolution Networks by Data Characteristic

ClassSR (CVPR2021) ClassSR: A General Framework to Accelerate Super-Resolution Networks by Data Characteristic Paper Authors: Xiangtao Kong, Hengyuan

Xiangtao Kong 308 Jan 05, 2023
Code for the Shortformer model, from the paper by Ofir Press, Noah A. Smith and Mike Lewis.

Shortformer This repository contains the code and the final checkpoint of the Shortformer model. This file explains how to run our experiments on the

Ofir Press 138 Apr 15, 2022
PyTorch code of "SLAPS: Self-Supervision Improves Structure Learning for Graph Neural Networks"

SLAPS-GNN This repo contains the implementation of the model proposed in SLAPS: Self-Supervision Improves Structure Learning for Graph Neural Networks

60 Dec 22, 2022
Unsupervised Image Generation with Infinite Generative Adversarial Networks

Unsupervised Image Generation with Infinite Generative Adversarial Networks Here is the implementation of MICGANs using DCGAN architecture on MNIST da

16 Dec 24, 2021
Generalized Data Weighting via Class-level Gradient Manipulation

Generalized Data Weighting via Class-level Gradient Manipulation This repository is the official implementation of Generalized Data Weighting via Clas

18 Nov 12, 2022
Tensors and Dynamic neural networks in Python with strong GPU acceleration

PyTorch is a Python package that provides two high-level features: Tensor computation (like NumPy) with strong GPU acceleration Deep neural networks b

61.4k Jan 04, 2023
Tensorflow implementation for Self-supervised Graph Learning for Recommendation

If the compilation is successful, the evaluator of cpp implementation will be called automatically. Otherwise, the evaluator of python implementation will be called.

152 Jan 07, 2023
SMCA replication There are no extra compiled components in SMCA DETR and package dependencies are minimal

Usage There are no extra compiled components in SMCA DETR and package dependencies are minimal, so the code is very simple to use. We provide instruct

22 May 06, 2022