Update README.md

README.md

@@ -9,7 +9,7 @@ The average relative rms error of 21cmLSTM is 0.22% when trained and tested on t
 The average evaluation speed of 21cmLSTM is 46 ms when emulating one signal at a time and using GPU acceleration (see Section 3.2 of DJ+24). This timing test was done using the eval_21cmGEM.py script and the computational resources stated in Section 2.3 of DJ+24. We used 21cmLSTM to fit mock data of the global 21 cm signal and showed that it can sufficiently exploit even outstandingly optimisic measurements of the signal (i.e., with 5 mK noise) and obtain unbiased Bayesian posterior constraints (see Section 3.2 of DJ+24). 21cmLSTM is thus useful for performing efficient and accurate multi-parameter inference of upcoming 21 cm experiments, which may require jointly fitting various complementary data sets or summary statistics (e.g., [Dorigo Jones et al. 2023](https://ui.adsabs.harvard.edu/abs/2023ApJ...959...49D/abstract); [Breitman et al. 2024](https://ui.adsabs.harvard.edu/abs/2024MNRAS.527.9833B/abstract)).
 
 ### GitHub description
-21cmLSTM is free to use on the MIT open source license. This GitHub contains the full code for training and running the emulator, including data preprocessing, network training, and model evaluation, to facilitate the community in retraining 21cmLSTM on other data sets and models. The sample notebook 'evaluate.ipynb' evaluates the provided representative instances of 21cmLSTM trained on either the 21cmGEM or ARES model data sets and prints accuracy metrics and plots, and the notebook 'train.ipynb' trains a new instance of 21cmLSTM on the same 21cmGEM data set used in DJ+24 (as well as used for other emulators, see above). The instance of 21cmLSTM trained on 21cmGEM included in this repository is the same one used for Bayesian analyses in DJ+24 and has test set mean rms error of 0.20%, and the instance of 21cmLSTM trained on ARES is a representative trial with mean emulation error of 0.38%.
+21cmLSTM is free to use on the MIT open source license. This GitHub contains the full code for training and running the emulator, including data preprocessing, network training, and model evaluation, to facilitate the community in retraining 21cmLSTM on other data sets and models. The sample notebook 'evaluate.ipynb' evaluates the provided representative trained instances of 21cmLSTM on the 21cmGEM and ARES model data sets and prints accuracy metrics, and the notebook 'train.ipynb' trains a new instance of 21cmLSTM on the 21cmGEM set with the same training settings described in Section 2.3 of DJ+24. The instance of 21cmLSTM trained on 21cmGEM included in this repository is the same one used for Bayesian analyses in DJ+24 and has test set mean rms error of 0.20%, and the instance of 21cmLSTM trained on ARES is a representative trial with mean emulation error of 0.38%.
 
 ### Contact; papers to reference
 Please reach out to me at johnny.dorigojones@colorado.edu about any comments, questions, or contributions (can also open an issue or make a pull request). Please reference Dorigo Jones et al. 2024 and provide a link to this GitHub repository if you utilize this work or emulator in any way, and [Dorigo Jones et al. 2023](https://ui.adsabs.harvard.edu/abs/2023ApJ...959...49D/abstract) regarding posterior emulation bias or if you perform joint fits using 21cmLSTM.