An implementation of Equivariant e2 convolutional kernals into a convolutional self attention network, applied to radio astronomy data.

Last update: Nov 09, 2021

Related tags

Deep Learning EquivariantSelfAttention

Overview

EquivariantSelfAttention

An implementation of Equivariant e2 convolutional kernals into a convolutional self attention network, applied to radio astronomy data.

This work extends and builds on previous work:

FR Classification using e2cnn, whose code is here.
Attention for FR Classificaion, whose code is here. This work makes extensive use of the e2cnn package: e2cnn.

Imports for General Use

import os
import torch
import e2cnn
import pandas as pd
import numpy as np
import configparser as ConfigParser

import utils

Loading a network:

To load a network import the network files, read in a config file, and load the network with the config file parameters:

from networks import AGRadGalNet, VanillaLeNet, testNet, DNSteerableLeNet, DNSteerableAGRadGalNet
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

config      = ConfigParser.ConfigParser(allow_no_value=True)
config_path = "configs/CONFIG_NAME"
config.read(config_path)

net = locals()[config['model']['base']](**config['model']).to(device)

Alternatively, if you know which architecture you want to use, only import that and load in the config parameters (i.e. without using locals()):

from networks import DNSteerableAGRadGalNet
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

config      = ConfigParser.ConfigParser(allow_no_value=True)
config_path = "configs/CONFIG_NAME"
config.read(config_path)

net = DNSteerableAGRadGalNet(**config['model']).to(device)

Loading a model:

To load the best performing model (according to early stopping) from a given model training session:

path_supliment = config['data']['augment']+'/'
model = utils.utils.load_model(config, load_model='best', device=device, path_supliment=path_supliment)

Testing a model:

Follow the example of evaluation.ipynb for model evaluation.

Defining a model:

To define your own model simply create a new config which follows the examples provided in configs.

Training a model:

To train a model use the train.py using your config file:

python3.8 train.py --config YOUR_CONFIG_NAME.cfg >& logs/YOUR_CONFIG_NAME.log

An implementation of Equivariant e2 convolutional kernals into a convolutional self attention network, applied to radio astronomy data.

Related tags

Overview

EquivariantSelfAttention

Imports for General Use

Loading a network:

Loading a model:

Testing a model:

Defining a model:

Training a model:

Owner

PyTorch code for Composing Partial Differential Equations with Physics-Aware Neural Networks

Only works with the dashboard version / branch of jesse

Source code for GNN-LSPE (Graph Neural Networks with Learnable Structural and Positional Representations)

Code for Multinomial Diffusion

Code to reproduce experiments in the paper "Explainability Requires Interactivity".

Official implementation of the RAVE model: a Realtime Audio Variational autoEncoder

[ICLR 2021] Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularization

MAterial del programa Misión TIC 2022

Code for the paper "Training GANs with Stronger Augmentations via Contrastive Discriminator" (ICLR 2021)

Multiview 3D object detection on MultiviewC dataset through moft3d.

A decent AI that solves daily Wordle puzzles. Works with different websites with similar wordlists,.

Pytorch implementation for the Temporal and Object Quantification Networks (TOQ-Nets).

Intelligent Video Analytics toolkit based on different inference backends.

Jremesh-tools - Blender addon for quad remeshing

Pre-Training 3D Point Cloud Transformers with Masked Point Modeling

Lex Rosetta: Transfer of Predictive Models Across Languages, Jurisdictions, and Legal Domains

Code and description for my BSc Project, September 2021

Face Mesh is a face geometry solution that estimates 468 3D face landmarks in real-time even on mobile devices

Official implementation of the MM'21 paper Constrained Graphic Layout Generation via Latent Optimization

Applying curriculum to meta-learning for few shot classification