%global _empty_manifest_terminate_build 0 Name: python-IMNN Version: 0.3.2 Release: 1 Summary: Using neural networks to extract sufficient statistics from data by maximising the Fisher information License: MIT License URL: https://bitbucket.org/tomcharnock/imnn.git Source0: https://mirrors.nju.edu.cn/pypi/web/packages/c7/2f/ec68a74f305b700245f3c6b6fb08870937a017b778fe8fa79bf2e895596f/IMNN-0.3.2.tar.gz BuildArch: noarch Requires: python3-jax Requires: python3-jaxlib Requires: python3-tensorflow Requires: python3-tqdm Requires: python3-tensorflow-probability[jax] Requires: python3-numpy Requires: python3-scipy Requires: python3-matplotlib %description |doc| |pypi| |bit| |git| |doi| |zen| The IMNN is a statistical method for transformation and compression of data from complex, high-dimensional distributions down to the number of physical parameters in the model which generates that data. It is asymptotically lossless in terms of information about the physical parameters. The method uses neural networks as a backbone for the transformation although any parameterised transformation with enough flexibility to map the data could be used. Using simulations generated from the model the Fisher information (for a Gaussian distribution with parameter independent covariance) is calculated from the output of the transformation and its log determinant is maximised under the condition that the covariance of the outputted transformed simulations are approximately constant and approach the identity matrix. Since the Fisher information only needs to be evaluated at a single fiducial parameter value it is exceptionally cheap to fit in comparison to other types of neural networks at the expense that the information is extracted from the data optimally about that fiducial parameter value and that the gradients of the simulations are needed (although this can be done numerically). The ideal situation for performing inference is to fit an IMNN at a fiducial choice of parameters, evaluate the quasi-maximum-likelihood estimate of the parameters for the target data of choice and retrain a new IMNN at this estimate (still very cheap). This can be repeated iteratively until the IMNN is being trained at the maximum likelihood values of the parameters which will be most sensitive to the features in the data at that point and allow for the most precise constraints. Note that the parameter estimates from the IMNN are not intended to be indicative of unbiased parameter estimates if the fiducial parameter choice is far from the maximum likelihood parameter values from some target. Instead they are meant to be used in a likelihood-free (simulation-based) inference scenario. %package -n python3-IMNN Summary: Using neural networks to extract sufficient statistics from data by maximising the Fisher information Provides: python-IMNN BuildRequires: python3-devel BuildRequires: python3-setuptools BuildRequires: python3-pip %description -n python3-IMNN |doc| |pypi| |bit| |git| |doi| |zen| The IMNN is a statistical method for transformation and compression of data from complex, high-dimensional distributions down to the number of physical parameters in the model which generates that data. It is asymptotically lossless in terms of information about the physical parameters. The method uses neural networks as a backbone for the transformation although any parameterised transformation with enough flexibility to map the data could be used. Using simulations generated from the model the Fisher information (for a Gaussian distribution with parameter independent covariance) is calculated from the output of the transformation and its log determinant is maximised under the condition that the covariance of the outputted transformed simulations are approximately constant and approach the identity matrix. Since the Fisher information only needs to be evaluated at a single fiducial parameter value it is exceptionally cheap to fit in comparison to other types of neural networks at the expense that the information is extracted from the data optimally about that fiducial parameter value and that the gradients of the simulations are needed (although this can be done numerically). The ideal situation for performing inference is to fit an IMNN at a fiducial choice of parameters, evaluate the quasi-maximum-likelihood estimate of the parameters for the target data of choice and retrain a new IMNN at this estimate (still very cheap). This can be repeated iteratively until the IMNN is being trained at the maximum likelihood values of the parameters which will be most sensitive to the features in the data at that point and allow for the most precise constraints. Note that the parameter estimates from the IMNN are not intended to be indicative of unbiased parameter estimates if the fiducial parameter choice is far from the maximum likelihood parameter values from some target. Instead they are meant to be used in a likelihood-free (simulation-based) inference scenario. %package help Summary: Development documents and examples for IMNN Provides: python3-IMNN-doc %description help |doc| |pypi| |bit| |git| |doi| |zen| The IMNN is a statistical method for transformation and compression of data from complex, high-dimensional distributions down to the number of physical parameters in the model which generates that data. It is asymptotically lossless in terms of information about the physical parameters. The method uses neural networks as a backbone for the transformation although any parameterised transformation with enough flexibility to map the data could be used. Using simulations generated from the model the Fisher information (for a Gaussian distribution with parameter independent covariance) is calculated from the output of the transformation and its log determinant is maximised under the condition that the covariance of the outputted transformed simulations are approximately constant and approach the identity matrix. Since the Fisher information only needs to be evaluated at a single fiducial parameter value it is exceptionally cheap to fit in comparison to other types of neural networks at the expense that the information is extracted from the data optimally about that fiducial parameter value and that the gradients of the simulations are needed (although this can be done numerically). The ideal situation for performing inference is to fit an IMNN at a fiducial choice of parameters, evaluate the quasi-maximum-likelihood estimate of the parameters for the target data of choice and retrain a new IMNN at this estimate (still very cheap). This can be repeated iteratively until the IMNN is being trained at the maximum likelihood values of the parameters which will be most sensitive to the features in the data at that point and allow for the most precise constraints. Note that the parameter estimates from the IMNN are not intended to be indicative of unbiased parameter estimates if the fiducial parameter choice is far from the maximum likelihood parameter values from some target. Instead they are meant to be used in a likelihood-free (simulation-based) inference scenario. %prep %autosetup -n IMNN-0.3.2 %build %py3_build %install %py3_install install -d -m755 %{buildroot}/%{_pkgdocdir} if [ -d doc ]; then cp -arf doc %{buildroot}/%{_pkgdocdir}; fi if [ -d docs ]; then cp -arf docs %{buildroot}/%{_pkgdocdir}; fi if [ -d example ]; then cp -arf example %{buildroot}/%{_pkgdocdir}; fi if [ -d examples ]; then cp -arf examples %{buildroot}/%{_pkgdocdir}; fi pushd %{buildroot} if [ -d usr/lib ]; then find usr/lib -type f -printf "/%h/%f\n" >> filelist.lst fi if [ -d usr/lib64 ]; then find usr/lib64 -type f -printf "/%h/%f\n" >> filelist.lst fi if [ -d usr/bin ]; then find usr/bin -type f -printf "/%h/%f\n" >> filelist.lst fi if [ -d usr/sbin ]; then find usr/sbin -type f -printf "/%h/%f\n" >> filelist.lst fi touch doclist.lst if [ -d usr/share/man ]; then find usr/share/man -type f -printf "/%h/%f.gz\n" >> doclist.lst fi popd mv %{buildroot}/filelist.lst . mv %{buildroot}/doclist.lst . %files -n python3-IMNN -f filelist.lst %dir %{python3_sitelib}/* %files help -f doclist.lst %{_docdir}/* %changelog * Mon May 29 2023 Python_Bot - 0.3.2-1 - Package Spec generated