v0.3.0

Radionets 0.3.0 (2023-07-20) (Second Paper Release)

API Changes

Bug Fixes

• Fix loading of correct sampling file [#145]

• • calculate nomalization only on non-zero pixels
  • fix typo in rescaling operation [#149]

• fixed sampling for images displayed in real and imaginary part [#152]

New Features

• enabled training and evaluation of half sized images (for 128 pixel images) [#140]

• Add naming to save path, so that the files do not get overwritten as easily [#144]

• • Add normalization callback with two different techniques
  • Update plotting routines for real/imag images
  • Update evaluate_area and evaluate_ms_ssim for half images
  • Add evaluate_ms_ssim for sampled images [#146]

• add evaluation of intensity via peak flux and integrated flux comparison [#150]

• • centered bin on 1 for histogram evaluation plots
  • added color to legend [#151]

• add prettier labels and descriptions to plots [#152]

Maintenance

• Deleted unusable functions for new source types

• Deleted unused hardcoded scaling [#140]

• • add masked loss functions
  • sort bundles in simulations
  • minor adjustments in plotting scripts [#141]

• consistent use of batch_size [#142]

• • Add the model name to predictions and sampling file
  • Delete unnecessary pad_unsqueeze function
  • Add amp_phase keyword to sample_images
  • Fix deprecation warning in sampling.py
  • Add image size to test_evaluation.py routines [#146]

• Outsource preprocessing steps in train_inspection.py [#148]

• Remove unused norm_path from all instances [#153]

• • Deleted cropping
  • updated colorbar label
  • removed source_list argument [#154]

Refactoring and Optimization

• Optimize evaluation.utils.trunc_rvs with numba, providing functions compiled for cpu and parallel cpu computation. [#143]

v0.2.0

Radionets 0.2.0 (2023-01-31)

API Changes

• train on half-sized images and applying symmetry afterward is a backward incompatible change models trained with early versions of radionets are not supported anymore #140

Bug Fixes

• fixed sampling of test data set fixed same indices for plots #140

New Features

• enabled training and evaluation of half sized images (for 128 pixel images) #140

Maintenance

• Deleted unusable functions for new source types Deleted unused hard coded scaling #140

Refactoring and Optimization

v0.1.18

Radionets 0.1.18 (2023-01-30)

API Changes

Bug Fixes

New Features

• added creation of uncertainty plots
  changed creation and saving/reading of predictions to dicts
  prediction dicts have 3 or 4 entries depending on uncertainty
  added scaled option to get_ifft
  created new dataset class for sampled images
  created option for sampling and saving the whole test dataset
  updated and wrote new tests #129

Maintenance

• Add and enable towncrier in CI. #130

• publish radionets on pypi #134

• Update README, use figures from the paper, minor text adjustments #136

Refactoring and Optimization

Initial publish on pypi

First published version of radionets.

PhD thesis K. Schmidt

This state was used to create the Ph.D. thesis of K. Schmidt. From this point, we plan backward incompatible changes.

Overhaul of the structure

Restructuring of the whole project. Different tasks are new available as command line executables.

Felix Geyers Master Thesis

State of the framework used for the Master Thesis from Felix Geyer submitted on 30 September 2020.

Feasibility Study without Sampling

Testing Fourier transformation with CNN using all available frequencies:

• Create toy data with MNIST
• Image size 64 x 64 pixel
• Two different models
• Training framework based on fast.ai course v3 part 2 (first half only)