gp-net is a regression tool for predicting the optical properties of materials,
and estimates the uncertainties on these predictions for the purpose
of active learning.
- Uncertainty Quantification
- Train-test split
- k-fold cross-validation
- Pool-based sampling Active Learning
- Entropy-based sampling
- Random-based sampling
usage: gp-net.py [-h] [-checkdata] [-ltype LTYPE] [-nomeg] [-noactive]
[-samp SAMP] [-cycle CYCLE CYCLE] [-repeat] [-q QUAN]
[-stop STOP] [-data DATA [DATA ...]] [-key KEY [KEY ...]]
[-frac FRAC [FRAC ...]] [-include] [-nsplit NSPLIT]
[-epochs EPOCHS] [-batch BATCH] [-bond BOND] [-nfeat NFEAT]
[-cutoff CUTOFF] [-width WIDTH] [-prev] [-layer LAYER]
[-ndims NDIMS] [-p PERP] [-niters NITERS] [-rate RATE]
[-amp AMP] [-length LENGTH]
[-maxiters MAXITERS [MAXITERS ...]]
Uncertainty quantification in neural networks.
optional arguments:
-h, --help show this help message and exit
-checkdata Check number of entries in the dataset. [default:
False]
-ltype LTYPE Display the layers in a fitted MEGNet model.
-nomeg Do not train with MEGNet. [default: False]
-noactive Don't do active learning [default: False]
-samp SAMP Type of sampling for active learning. Use random or
entropy [default: entropy]
-cycle CYCLE CYCLE Number of structures to sample and maximum number of
times to sample separated by spaces for active
learning. [default: 1 5]
-repeat MEGNet train and pre-process activations in each
active learning cycle [default: False]
-q QUAN, --quan QUAN Quantity of data for norepeat active learning
[default: 1000]
-stop STOP Maximum fraction of test set required for active
learning [default: 0.1]
-data DATA [DATA ...]
Input dataset(s). Multiple datasets can be passed, one
per optical property of interest. [No default]
-key KEY [KEY ...] API key for data download and the optical properties
of interest, separated by spaces. For MEGNet users
only. [eg. Key band_gap formation_energy_per_atom
e_above_hull]
-frac FRAC [FRAC ...]
Fraction of data for training and testing separated by
spaces for train-test split and k-fold cross-
validation. Fraction of data for training, and
fraction of training data for validation in repeat
active learning. For norepeat active learning, single
input as the fraction of the training data for
validation. [default: 0.3]
-include Include zero optical property values in the MEGNet
training and/or Gaussian process analysis. [default:
False]
-nsplit NSPLIT Number of training set splits for k-fold cross-
validation. [default: 1 i.e no cross-validation]
-epochs EPOCHS Epochs. [default: 0 ie. Perform no training with
MEGNet]
-batch BATCH Batch size for training with MEGNet or CNN. [default:
256]
-bond BOND MEGNet feature bond. [default: 10]
-nfeat NFEAT MEGNet feature global. [default: 2]
-cutoff CUTOFF, --cutoff CUTOFF
MEGNet radial cutoff. [default: 5]
-width WIDTH, --width WIDTH
MEGNet gaussian width. [default: 0.5]
-prev Use a pre-trained MEGNet model during training with
MEGNet. [default: False]
-layer LAYER MEGNet fitted model layer to analyse. [default:
readout_0 i.e 32 dense layer]
-ndims NDIMS Dimensions of embedded space. 0 => Do not preprocess
activations, 1 => scale activations to 0, 1 range, 2
or 3 => Reduce dimensions of activations with tSNE.
[default: 0]
-ndims NDIMS Dimensions of embedded space. 0 => scale activations
in 0,1 range 2 or 3 => Reduce dimensions of
activations with tSNE. [default: 0]
-p PERP, --perp PERP Perplexity value to use in dimension reduction with
tSNE. [default: 150]
-niters NITERS Number of iterations for optimisation in tSNE.
[default: 1000]
-rate RATE Adam optimizer Learning rate. [default: 0.01]
-amp AMP Amplitude of the GP kernel. [default: 1.0]
-length LENGTH The length scale of the GP kernel. [default: 1.0]
-maxiters MAXITERS [MAXITERS ...]
Maximum iterations for optimising GP hyperparameters.
For k-fold cross-validation, two inputs are required -
one for training per fold and the other for training
using train-test split. For active learning and train-
test split, a single input is required. [default: 0
i.e no GP training]
Please see the wiki page for description
of all the features of gp-net. If your questions are not answered in the wiki,
please contact us by email. If you have found a bug in any of the tools, please
submit a ticket and we will attend to it.
Cite this repo as follows:
@misc{injector:2021,
title = {Injector-Surrogates: Machine Learning Surrogate Models for Particle Accelerator Injector Profile Simulation},
author = {Johannes Allotey, Keith T. Butler, Jeyan Thiyagalingam},
url = {https://github.com/keeeto/gp-net/},
year = {2021}
}
This work was partially supported by wave 1 of the UKRI Strategic Priorities Fund under the EPSRC (Grant No. EP/T001569/1), particularly the “AI for Science” theme within that grant and The Alan Turing Institute. The ML models were trained using computing resources provided by STFC Scientific Computing Department’s SCARF cluster and the PEARL cluster. We acknowledge support from STFC via the Data Intensive Centre for Doctoral Training (ST/P006779/1) and the University of Bristol.