|
Emulate whatever your ❤️ desires. Install with:pip install -v git+https://github.com/NoahSailer/EmulateIt |
Here's an example for generating training data:
cd path/to/EmulateIt/example
interact # launch interactive node
conda activate classyenv # enviornment for classy & mpi4py
srun -n 128 python matter_pk.py # generates (normalized) training data
The training data is saved as example/training-data_[inputs][outputs].npy in the
specified directory (in this case: example/). You can use the train-nn [train_dir/]
command within an enviornment where scikit-learn is installed to train the NN:
conda activate sklearn-env
export OMP_NUM_THREADS=128
srun -n 1 -c 128 train-nn ./
The weights are saved example/weights.json. To test the emulator against CLASS see
example/matter_pk_emu.py.
For now I'm using scikit-learn for training and evaluations. In the future we may want to have an
array of options to choose from, especially since these packages become obsolete fairly quickly.