Ctlearn

Major Features

Minor Improvements

Bug Fixes and Other Changes

* Improve installation instructions

Known Issues

* _apply_class_weights_ only supported for `CTLearn <= v0.5.1` and `dl1dh <= v0.10.2`. Please balance your dataset by hand beforehand. For `CTLearn v0.6.0` _apply_class_weights_ will be supported again and automatized.

GitHub actions and featuring of the LSTSiPM camera

Major Features
* Fixed GitHub actions 162 nietootein

Minor Improvements
* Added the LSTSiPM camera to the default config files 167
* Renamed the structure of CTLearn's core modules
* Get viewcone from difference of max and min stored in the file
* Improving docs and README
* Upgrade to dl1dh v0.10.10, ctapipe v0.19.0 , pyirf to v0.8, TensorFlow v2.9 & python 3.10

Bug Fixes and Other Changes

Known Issues
* Particle classification is not working with Multitask Learning models yet.

Output (dl2-like) format handling and creation of an IRF builder using pyirf 142
Store keras model in onnx format 143

Major Features
* Speeding up the output writing by pseudo-chunk processing of the keras predictions
* Clean up CNNRNN model via TimeDistributed layer
* Enable learning rate reducer (including early stopper) callback
* Automatised class label handling for multiple particle types
* Set cleaning from the command line via a flag. Therefore default models with cleaned images can be removed.

Minor Improvements
* Store only the best model checkpoints for validation metric
* Improve installation process of TF by removing cpu/gpu mode
* Upgrade supplementary scripts to the new output format
* Upgrade to dl1dh v0.10.7, ctapipe v0.15.0 & python 3.9

Bug Fixes and Other Changes

Known Issues
* Particle classification is not working with Multitask Learning models yet.
* There is some version incompatibility (numpy and numba) when trying to run on Wilkes-3.

Major upgrade to TensorFlow v2.8 137
Sphinx docs 139

Major Features
* Upgrade the models to the Keras API in order to run with TF v2.8
* Enable usage of multiple GPUs via `tf.distribute.MirroredStrategy`
* Make CTLearn user-friendly by allowing several analysis options (like reconstruction tasks, directories, telescope types/ids, quality cuts) to be set from command line. Therefore minimum information about the model has to be included in the config file. Default CTLearn models can be constructed from the command line via default config files shipped by the installation.
* Balance the data for the classification task by default
* Add Sphinx docs and additional code meta data nietootein

Minor Improvements
* ResNets can be constructed with the SingleCNN model.
* Store Only the best model checkpoints
* Model architecture & matrices can be plotted automatically
* Upgrade to dl1dh v0.10.5, ctapipe v0.12.0 & python 3.8

Bug Fixes and Other Changes

Known Issues
* Particle classification is not working with Multitask Learning models yet.
* There is some version incompatibility (numpy and numba) when trying to run on Wilkes-3.

ctapipe-stage1 migration

Major Features
* Support of the official CTA stage 1 files (the dl1dh data format is still supported but will be deprecated in the future)
* Upgrade to dl1dh v0.10.0 and ctapipe v0.10.5
* Added ResNet-RNN model
* New feature: Freeze the backbone in deep stereo models like the ResNet-RNN
* pypi installation for CTLearn

Minor Improvements
* Add CTA and MAGIC example files
* Improve input file handling in predict mode to process real data in standard convention

Bug Fixes and Other Changes