History¶
v1.1 series¶
v1.1.0 (xx xxx 20xx)
Added models :
ApogeeKplerEchelle
Input data can now be a dict, such as
nn.train({'input': input_data, 'input': aux_input_data}, {'output': labels, 'output_aux': aux_labels})
Added numerical integrator for NeuralODE
Fully compatible with Tensorflow 2
Model’s functions should be much faster due to using Tensorflow v2 eager execution (see: https://github.com/tensorflow/tensorflow/issues/33024#issuecomment-551184305)
Improved continuous integration testing, now actually test model learn properly with real world data instead of checking no syntax error with random data
Deprecated support for all Tensorflow 1.x
Dropped optional Keras support, now depends on Tensorflow only
Tested with Tensorflow 2.4.x
Incompatible to Tensorflow 1.x and <=2.2 due to necessary changes for Tensorflow eager execution API
v1.0 series¶
v1.0.1 (5 March 2019)
This release mainly targeted to the paper Simultaneous calibration of spectro-photometric distances and the Gaia DR2 parallax zero-point offset with deep learning
available at
[arXiv:1902.08634]
[ADS]
Documentation for this version is available at https://astronn.readthedocs.io/en/v1.0.1/
Better and faster with IPython tab auto-completion
Added models :
ApogeeDR14GaiaDR2BCNN
Improved data pipeline to generate data for NNs
Tested with Tensorflow 1.11.0/1.12.0/1.13.1 and Keras 2.2.0/2.2.4
v1.0.0 (16 August 2018)
This is the first release of astroNN. This release mainly targeted to the paper Deep learning of multi-element abundances from high-resolution spectroscopic data
available at
[arXiv:1804.08622]
[ADS]
Documentation for this version is available at https://astronn.readthedocs.io/en/v1.0.0/
Initial Release!!
Tested with Tensorflow 1.8.0/1.9.0 and Keras 2.2.0/2.2.2
Python 3.6 or above only
v0.0 series¶
v0.0.0 (13 October 2017)
First commit of astroNN on Github!!!