{"id":25971508,"url":"https://github.com/stefanradev93/BayesFlow","last_synced_at":"2025-03-05T00:01:49.804Z","repository":{"id":37817522,"uuid":"222209282","full_name":"bayesflow-org/bayesflow","owner":"bayesflow-org","description":"A Python library for amortized Bayesian workflows using generative neural networks.","archived":false,"fork":false,"pushed_at":"2024-11-01T12:32:55.000Z","size":187689,"stargazers_count":361,"open_issues_count":23,"forks_count":50,"subscribers_count":19,"default_branch":"master","last_synced_at":"2024-11-01T14:40:53.123Z","etag":null,"topics":["amortized-inference","bayesian-inference","deep-learning","invertible-networks","likelihood-free","model-comparison","parameter-estimation","simulation-based","uncertainty-quantification"],"latest_commit_sha":null,"homepage":"https://bayesflow.org/","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/bayesflow-org.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.rst","contributing":"CONTRIBUTING","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":"CITATION.cff","codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2019-11-17T06:53:05.000Z","updated_at":"2024-10-28T13:59:53.000Z","dependencies_parsed_at":"2023-02-19T01:16:07.463Z","dependency_job_id":"7c9e3aba-abe0-453d-96cb-4775b7af8278","html_url":"https://github.com/bayesflow-org/bayesflow","commit_stats":{"total_commits":935,"total_committers":19,"mean_commits":49.21052631578947,"dds":0.2823529411764706,"last_synced_commit":"fda44790cd9a32111d88b1461be7e4e116c1cb33"},"previous_names":["bayesflow-org/bayesflow","stefanradev93/bayesflow"],"tags_count":11,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bayesflow-org%2Fbayesflow","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bayesflow-org%2Fbayesflow/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bayesflow-org%2Fbayesflow/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bayesflow-org%2Fbayesflow/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/bayesflow-org","download_url":"https://codeload.github.com/bayesflow-org/bayesflow/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":241940552,"owners_count":20045881,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["amortized-inference","bayesian-inference","deep-learning","invertible-networks","likelihood-free","model-comparison","parameter-estimation","simulation-based","uncertainty-quantification"],"created_at":"2025-03-05T00:01:29.671Z","updated_at":"2025-03-05T00:01:49.764Z","avatar_url":"https://github.com/bayesflow-org.png","language":"Python","funding_links":[],"categories":["Code Packages and Benchmarks"],"sub_categories":[],"readme":"# BayesFlow \u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/bayesflow_hex.png?raw=true\" align=\"right\" width=20% height=20% /\u003e\n\n[![Actions Status](https://github.com/stefanradev93/bayesflow/workflows/Tests/badge.svg)](https://github.com/stefanradev93/bayesflow/actions)\n[![License: MIT](https://img.shields.io/badge/License-MIT-red.svg)](https://opensource.org/licenses/MIT)\n[![DOI](https://joss.theoj.org/papers/10.21105/joss.05702/status.svg)](https://doi.org/10.21105/joss.05702)\n[![contributions welcome](https://img.shields.io/badge/contributions-welcome-brightgreen.svg?style=flat)](https://github.com/dwyl/esta/issues)\n\n\nWelcome to our BayesFlow library for efficient simulation-based Bayesian workflows! Our library enables users to create specialized neural networks for *amortized Bayesian inference*, which repay users with rapid statistical inference after a potentially longer simulation-based training phase.\n\n\u003e 🚨 **Attention, new users!** 🚨\n\u003e This is the `master` branch of BayesFlow, which only supports TensorFlow. The `dev` branch contains the new BayesFlow version 2.0 that fully builds on Keras 3. This means you can choose your backend (PyTorch, JAX, TensorFlow) and have full flexibility. We are actively working on this new BayesFlow version and will merge it into the `master` branch once all features are implemented.\n\u003e [Click here (Link)](https://github.com/stefanradev93/BayesFlow/tree/dev) to get to the dev branch and see the latest installation instructions.\n\n\nFor starters, check out some of our walk-through notebooks:\n\n1. [Quickstart amortized posterior estimation](examples/Intro_Amortized_Posterior_Estimation.ipynb)\n2. [Tackling strange bimodal distributions](examples/TwoMoons_Bimodal_Posterior.ipynb)\n3. [Detecting model misspecification in posterior inference](examples/Model_Misspecification.ipynb)\n4. [Principled Bayesian workflow for cognitive models](examples/LCA_Model_Posterior_Estimation.ipynb)\n5. [Posterior estimation for ODEs](examples/Linear_ODE_system.ipynb)\n6. [Posterior estimation for SIR-like models](examples/Covid19_Initial_Posterior_Estimation.ipynb)\n7. [Model comparison for cognitive models](examples/Model_Comparison_MPT.ipynb)\n8. [Hierarchical model comparison for cognitive models](examples/Hierarchical_Model_Comparison_MPT.ipynb)\n9. [Two level model with sequential observations](examples/Two_Level_Sequential_Model.ipynb)\n\n## Documentation \\\u0026 Help\n\nThe project documentation is available at \u003chttps://bayesflow.org\u003e. Please use the [BayesFlow Forums](https://discuss.bayesflow.org/) for any BayesFlow-related questions and discussions, and [GitHub Issues](https://github.com/stefanradev93/BayesFlow/issues) for bug reports and feature requests.\n\n## Installation\n\nSee [INSTALL.rst](INSTALL.rst) for installation instructions.\n\n## Conceptual Overview\n\nA cornerstone idea of amortized Bayesian inference is to employ generative\nneural networks for parameter estimation, model comparison, and model validation\nwhen working with intractable simulators whose behavior as a whole is too\ncomplex to be described analytically. The figure below presents a higher-level\noverview of neurally bootstrapped Bayesian inference.\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/high_level_framework.png?raw=true\" width=80% height=80%\u003e\n\n## Getting Started: Parameter Estimation\n\nThe core functionality of BayesFlow is amortized Bayesian posterior estimation, as described in our paper:\n\nRadev, S. T., Mertens, U. K., Voss, A., Ardizzone, L., \u0026 Köthe, U. (2020).\nBayesFlow: Learning complex stochastic models with invertible neural networks.\n\u003cem\u003eIEEE Transactions on Neural Networks and Learning Systems\u003c/em\u003e, available\nfor free at: https://arxiv.org/abs/2003.06281.\n\nHowever, since then, we have substantially extended the BayesFlow library such that\nit is now much more general and cleaner than what we describe in the above paper.\n\n### Minimal Example\n\n```python\nimport numpy as np\nimport bayesflow as bf\n```\n\nTo introduce you to the basic workflow of the library, let's consider\na simple 2D Gaussian model, from which we want to obtain\nposterior inference.  We assume a Gaussian simulator (likelihood)\nand a Gaussian prior for the means of the two components,\nwhich are our only model parameters in this example:\n\n```python\ndef simulator(theta, n_obs=50, scale=1.0):\n    return np.random.default_rng().normal(loc=theta, scale=scale, size=(n_obs, theta.shape[0]))\n\ndef prior(D=2, mu=0., sigma=1.0):\n    return np.random.default_rng().normal(loc=mu, scale=sigma, size=D)\n```\n\nThen, we connect the `prior` with the `simulator` using a `GenerativeModel` wrapper:\n\n```python\ngenerative_model = bf.simulation.GenerativeModel(prior, simulator, simulator_is_batched=False)\n```\n\nNext, we create our BayesFlow setup consisting of a summary and an inference network:\n\n```python\nsummary_net = bf.networks.SetTransformer(input_dim=2)\ninference_net = bf.networks.InvertibleNetwork(num_params=2)\namortized_posterior = bf.amortizers.AmortizedPosterior(inference_net, summary_net)\n```\n\nFinally, we connect the networks with the generative model via a `Trainer` instance:\n\n```python\ntrainer = bf.trainers.Trainer(amortizer=amortized_posterior, generative_model=generative_model)\n```\n\nWe are now ready to train an amortized posterior approximator. For instance,\nto run online training, we simply call:\n\n```python\nlosses = trainer.train_online(epochs=10, iterations_per_epoch=1000, batch_size=32)\n```\n\nPrior to inference, we can use simulation-based calibration (SBC,\nhttps://arxiv.org/abs/1804.06788) to check the computational faithfulness of\nthe model-amortizer combination on unseen simulations:\n\n```python\n# Generate 500 new simulated data sets\nnew_sims = trainer.configurator(generative_model(500))\n\n# Obtain 100 posterior draws per data set instantly\nposterior_draws = amortized_posterior.sample(new_sims, n_samples=100)\n\n# Diagnose calibration\nfig = bf.diagnostics.plot_sbc_histograms(posterior_draws, new_sims['parameters'])\n```\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/showcase_sbc.png?raw=true\" width=65% height=65%\u003e\n\nThe histograms are roughly uniform and lie within the expected range for\nwell-calibrated inference algorithms as indicated by the shaded gray areas.\nAccordingly, our neural approximator seems to have converged to the intended target.\n\nAs you can see, amortized inference on new (real or simulated) data is easy and fast.\nWe can obtain further 5000 posterior draws per simulated data set and quickly inspect\nhow well the model can recover its parameters across the entire *prior predictive distribution*.\n\n\n```python\nposterior_draws = amortized_posterior.sample(new_sims, n_samples=5000)\nfig = bf.diagnostics.plot_recovery(posterior_draws, new_sims['parameters'])\n```\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/showcase_recovery.png?raw=true\" width=65% height=65%\u003e\n\nFor any individual data set, we can also compare the parameters' posteriors with\ntheir corresponding priors:\n\n```python\nfig = bf.diagnostics.plot_posterior_2d(posterior_draws[0], prior=generative_model.prior)\n```\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/showcase_posterior.png?raw=true\" width=45% height=45%\u003e\n\nWe see clearly how the posterior shrinks relative to the prior for both\nmodel parameters as a result of conditioning on the data.\n\n### References and Further Reading\n\n- Radev, S. T., Mertens, U. K., Voss, A., Ardizzone, L., \u0026 Köthe, U. (2020).\nBayesFlow: Learning complex stochastic models with invertible neural networks.\n\u003cem\u003eIEEE Transactions on Neural Networks and Learning Systems, 33(4)\u003c/em\u003e, 1452-1466.\n\n- Radev, S. T., Graw, F., Chen, S., Mutters, N. T., Eichel, V. M., Bärnighausen, T., \u0026 Köthe, U. (2021).\nOutbreakFlow: Model-based Bayesian inference of disease outbreak dynamics with invertible neural networks and its application to the COVID-19 pandemics in Germany. \u003cem\u003ePLoS Computational Biology, 17(10)\u003c/em\u003e, e1009472.\n\n- Bieringer, S., Butter, A., Heimel, T., Höche, S., Köthe, U., Plehn, T., \u0026 Radev, S. T. (2021).\nMeasuring QCD splittings with invertible networks. \u003cem\u003eSciPost Physics, 10(6)\u003c/em\u003e, 126.\n\n- von Krause, M., Radev, S. T., \u0026 Voss, A. (2022).\nMental speed is high until age 60 as revealed by analysis of over a million participants.\n\u003cem\u003eNature Human Behaviour, 6(5)\u003c/em\u003e, 700-708.\n\n## Model Misspecification\n\nWhat if we are dealing with misspecified models? That is, how faithful is our\namortized inference if the generative model is a poor representation of reality?\nA modified loss function optimizes the learned summary statistics towards a unit\nGaussian and reliably detects model misspecification during inference time.\n\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/examples/img/model_misspecification_amortized_sbi.png?raw=true\" width=100% height=100%\u003e\n\nIn order to use this method, you should only provide the `summary_loss_fun` argument\nto the `AmortizedPosterior` instance:\n\n```python\namortized_posterior = bf.amortizers.AmortizedPosterior(inference_net, summary_net, summary_loss_fun='MMD')\n```\n\nThe amortizer knows how to combine its losses and you can inspect the summary space for outliers during inference.\n\n### References and Further Reading\n\n- Schmitt, M., Bürkner P. C., Köthe U., \u0026 Radev S. T. (2022). Detecting Model\nMisspecification in Amortized Bayesian Inference with Neural Networks. \u003cem\u003eArXiv\npreprint\u003c/em\u003e, available for free at: https://arxiv.org/abs/2112.08866\n\n## Model Comparison\n\nBayesFlow can not only be used for parameter estimation, but also to perform approximate Bayesian model comparison via posterior model probabilities or Bayes factors.\nLet's extend the minimal example from before with a second model $M_2$ that we want to compare with our original model $M_1$:\n\n```python\ndef simulator(theta, n_obs=50, scale=1.0):\n    return np.random.default_rng().normal(loc=theta, scale=scale, size=(n_obs, theta.shape[0]))\n\ndef prior_m1(D=2, mu=0., sigma=1.0):\n    return np.random.default_rng().normal(loc=mu, scale=sigma, size=D)\n\ndef prior_m2(D=2, mu=2., sigma=1.0):\n    return np.random.default_rng().normal(loc=mu, scale=sigma, size=D)\n```\n\nFor the purpose of this illustration, the two toy models only differ with respect to their prior specification ($M_1: \\mu = 0, M_2: \\mu = 2$). We create both models as before and use a `MultiGenerativeModel` wrapper to combine them in a `meta_model`:\n\n```python\nmodel_m1 = bf.simulation.GenerativeModel(prior_m1, simulator, simulator_is_batched=False)\nmodel_m2 = bf.simulation.GenerativeModel(prior_m2, simulator, simulator_is_batched=False)\nmeta_model = bf.simulation.MultiGenerativeModel([model_m1, model_m2])\n```\n\nNext, we construct our neural network with a `PMPNetwork` for approximating posterior model probabilities:\n\n```python\nsummary_net = bf.networks.SetTransformer(input_dim=2)\nprobability_net = bf.networks.PMPNetwork(num_models=2)\namortized_bmc = bf.amortizers.AmortizedModelComparison(probability_net, summary_net)\n```\n\nWe combine all previous steps with a `Trainer` instance and train the neural approximator:\n\n```python\ntrainer = bf.trainers.Trainer(amortizer=amortized_bmc, generative_model=meta_model)\nlosses = trainer.train_online(epochs=3, iterations_per_epoch=100, batch_size=32)\n```\n\nLet's simulate data sets from our models to check our networks' performance:\n\n```python\nsims = trainer.configurator(meta_model(5000))\n```\n\nWhen feeding the data to our trained network, we almost immediately obtain posterior model probabilities for each of the 5000 data sets:\n\n```python\nmodel_probs = amortized_bmc.posterior_probs(sims)\n```\n\nHow good are these predicted probabilities in the closed world? We can have a look at the calibration:\n\n```python\ncal_curves = bf.diagnostics.plot_calibration_curves(sims[\"model_indices\"], model_probs)\n```\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/showcase_calibration_curves.png?raw=true\" width=65% height=65%\u003e\n\nOur approximator shows excellent calibration, with the calibration curve being closely aligned to the diagonal, an expected calibration error (ECE) near 0 and most predicted probabilities being certain of the model underlying a data set. We can further assess patterns of misclassification with a confusion matrix:\n\n```python\nconf_matrix = bf.diagnostics.plot_confusion_matrix(sims[\"model_indices\"], model_probs)\n```\n\n\u003cimg src=\"https://github.com/stefanradev93/BayesFlow/blob/master/img/showcase_confusion_matrix.png?raw=true\" width=44% height=44%\u003e\n\nFor the vast majority of simulated data sets, the \"true\" data-generating model is correctly identified. With these diagnostic results backing us up, we can proceed and apply our trained network to empirical data.\n\nBayesFlow is also able to conduct model comparison for hierarchical models. See this [tutorial notebook](examples/Hierarchical_Model_Comparison_MPT.ipynb) for an introduction to the associated workflow.\n\n### References and Further Reading\n\n- Radev S. T., D’Alessandro M., Mertens U. K., Voss A., Köthe U., \u0026 Bürkner P.\nC. (2021). Amortized Bayesian Model Comparison with Evidental Deep Learning.\n\u003cem\u003eIEEE Transactions on Neural Networks and Learning Systems\u003c/em\u003e.\ndoi:10.1109/TNNLS.2021.3124052 available for free at: https://arxiv.org/abs/2004.10629\n\n- Schmitt, M., Radev, S. T., \u0026 Bürkner, P. C. (2022). Meta-Uncertainty in\nBayesian Model Comparison. In \u003cem\u003eInternational Conference on Artificial Intelligence\nand Statistics\u003c/em\u003e, 11-29, PMLR, available for free at: https://arxiv.org/abs/2210.07278\n\n- Elsemüller, L., Schnuerch, M., Bürkner, P. C., \u0026 Radev, S. T. (2023). A Deep\nLearning Method for Comparing Bayesian Hierarchical Models. \u003cem\u003eArXiv preprint\u003c/em\u003e,\navailable for free at: https://arxiv.org/abs/2301.11873\n\n## Likelihood Emulation\n\nIn order to learn the exchangeable (i.e., permutation invariant) likelihood from the minimal example instead of the posterior, you may use the `AmortizedLikelihood` wrapper:\n\n```python\nlikelihood_net = bf.networks.InvertibleNetwork(num_params=2)\namortized_likelihood = bf.amortizers.AmortizedLikelihood(likelihood_net)\n```\n\nThis wrapper can interact with a `Trainer` instance in the same way as the `AmortizedPosterior`. Finally, you can also learn the likelihood and the posterior *simultaneously* by using the `AmortizedPosteriorLikelihood` wrapper and choosing your preferred training scheme:\n\n```python\njoint_amortizer = bf.amortizers.AmortizedPosteriorLikelihood(amortized_posterior, amortized_likelihood)\n```\n\nLearning both densities enables us to approximate marginal likelihoods or perform approximate leave-one-out cross-validation (LOO-CV) for prior or posterior predictive model comparison, respectively.\n\n### References and Further Reading\n\nRadev, S. T., Schmitt, M., Pratz, V., Picchini, U., Köthe, U., \u0026 Bürkner, P.-C. (2023).\nJANA: Jointly amortized neural approximation of complex Bayesian models.\n*Proceedings of the Thirty-Ninth Conference on Uncertainty in Artificial Intelligence, 216*, 1695-1706.\n([arXiv](https://arxiv.org/abs/2302.09125))([PMLR](https://proceedings.mlr.press/v216/radev23a.html))\n\n## Support\n\nThis project is currently managed by researchers from Rensselaer Polytechnic Institute, TU Dortmund University, and Heidelberg University. It is partially funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation, Project 528702768). The project is further supported by Germany's Excellence Strategy -- EXC-2075 - 390740016 (Stuttgart Cluster of Excellence SimTech) and EXC-2181 - 390900948 (Heidelberg Cluster of Excellence STRUCTURES), as well as the Informatics for Life initiative funded by the Klaus Tschira Foundation.\n\n## Citing BayesFlow\n\nYou can cite BayesFlow along the lines of:\n\n- We approximated the posterior with neural posterior estimation and learned summary statistics (NPE; Radev et al., 2020), as implemented in the BayesFlow software for amortized Bayesian workflows (Radev et al., 2023a).\n- We approximated the likelihood with neural likelihood estimation (NLE; Papamakarios et al., 2019) without hand-crafted summary statistics, as implemented in the BayesFlow software for amortized Bayesian workflows (Radev et al., 2023b).\n- We performed simultaneous posterior and likelihood estimation with jointly amortized neural approximation (JANA; Radev et al., 2023a), as implemented in the BayesFlow software for amortized Bayesian workflows (Radev et al., 2023b).\n\n1. Radev, S. T., Schmitt, M., Schumacher, L., Elsemüller, L., Pratz, V., Schälte, Y., Köthe, U., \u0026 Bürkner, P.-C. (2023a). BayesFlow: Amortized Bayesian workflows with neural networks. *The Journal of Open Source Software, 8(89)*, 5702.([arXiv](https://arxiv.org/abs/2306.16015))([JOSS](https://joss.theoj.org/papers/10.21105/joss.05702))\n2. Radev, S. T., Mertens, U. K., Voss, A., Ardizzone, L., Köthe, U. (2020). BayesFlow: Learning complex stochastic models with invertible neural networks. *IEEE Transactions on Neural Networks and Learning Systems, 33(4)*, 1452-1466. ([arXiv](https://arxiv.org/abs/2003.06281))([IEEE TNNLS](https://ieeexplore.ieee.org/document/9298920))\n3. Radev, S. T., Schmitt, M., Pratz, V., Picchini, U., Köthe, U., \u0026 Bürkner, P.-C. (2023b). JANA: Jointly amortized neural approximation of complex Bayesian models. *Proceedings of the Thirty-Ninth Conference on Uncertainty in Artificial Intelligence, 216*, 1695-1706. ([arXiv](https://arxiv.org/abs/2302.09125))([PMLR](https://proceedings.mlr.press/v216/radev23a.html))\n\n**BibTeX:**\n\n```\n@article{bayesflow_2023_software,\n  title = {{BayesFlow}: Amortized {B}ayesian workflows with neural networks},\n  author = {Radev, Stefan T. and Schmitt, Marvin and Schumacher, Lukas and Elsemüller, Lasse and Pratz, Valentin and Schälte, Yannik and Köthe, Ullrich and Bürkner, Paul-Christian},\n  journal = {Journal of Open Source Software},\n  volume = {8},\n  number = {89},\n  pages = {5702},\n  year = {2023}\n}\n\n@article{bayesflow_2020_original,\n  title = {{BayesFlow}: Learning complex stochastic models with invertible neural networks},\n  author = {Radev, Stefan T. and Mertens, Ulf K. and Voss, Andreas and Ardizzone, Lynton and K{\\\"o}the, Ullrich},\n  journal = {IEEE transactions on neural networks and learning systems},\n  volume = {33},\n  number = {4},\n  pages = {1452--1466},\n  year = {2020}\n}\n\n@inproceedings{bayesflow_2023_jana,\n  title = {{JANA}: Jointly amortized neural approximation of complex {B}ayesian models},\n  author = {Radev, Stefan T. and Schmitt, Marvin and Pratz, Valentin and Picchini, Umberto and K\\\"othe, Ullrich and B\\\"urkner, Paul-Christian},\n  booktitle = {Proceedings of the Thirty-Ninth Conference on Uncertainty in Artificial Intelligence},\n  pages = {1695--1706},\n  year = {2023},\n  volume = {216},\n  series = {Proceedings of Machine Learning Research},\n  publisher = {PMLR}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fstefanradev93%2FBayesFlow","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fstefanradev93%2FBayesFlow","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fstefanradev93%2FBayesFlow/lists"}