coexist


Namecoexist JSON
Version 0.3.2 PyPI version JSON
download
home_pagehttps://github.com/uob-positron-imaging-centre/Coexist
SummaryLearning simulation parameters from experimental data, from the micro to the macro, from the laptop to the cluster.
upload_time2023-06-11 02:52:57
maintainer
docs_urlNone
authorAndrei Leonard Nicusan <a.l.nicusan@bham.ac.uk>, Dominik Werner <d.werner.1@bham.ac.uk>, Jack Sykes <jas653@student.bham.ac.uk>
requires_python>=3.6
licenseGNU
keywords simulation optimization calibration parameter-estimation parameter-tuning physics-simulation computational-fluid-dynamics granular dem discrete-element-method medical-imaging
VCS
bugtrack_url
requirements No requirements were recorded.
Travis-CI No Travis.
coveralls test coverage No coveralls.
            [![PyPI version](https://badge.fury.io/py/coexist.svg)](https://badge.fury.io/py/coexist)
[![Documentation Status](https://readthedocs.org/projects/coexist/badge/?version=latest)](https://coexist.readthedocs.io/en/latest/?badge=latest)
[![CI Status](https://github.com/uob-positron-imaging-centre/Coexist/actions/workflows/ci.yml/badge.svg)](https://github.com/uob-positron-imaging-centre/Coexist/actions/workflows/ci.yml)
[![Language grade: Python](https://img.shields.io/lgtm/grade/python/g/uob-positron-imaging-centre/Coexist.svg?logo=lgtm&logoWidth=18)](https://lgtm.com/projects/g/uob-positron-imaging-centre/Coexist/context:python)
[![Colab example](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1p7OwnaYgENwK4DTn_6QETX4ajwVFeza0?usp=sharing)
[![License: GPL-3.0](https://img.shields.io/github/license/uob-positron-imaging-centre/Coexist?style=flat-square)](https://github.com/uob-positron-imaging-centre/Coexist)


# CoExSiST & ACCES
### Data-Driven Evolutionary Calibration & Optimisation of Simulations

A Python library for autonomously learning simulation parameters from experimental data, from the *micro* to the *macro*, from laptops to clusters. This is done using either of two closely related frameworks:

- **CoExSiST**: Coupled Experimental-Simulational Study Tool.
- **ACCES**: Autonomous Characterisation and Calibration via Evolutionary Simulation. 

Both libraries learn a given set of free parameters, such that an experiment is synchronised with an equivalent simulation; this synchronisation is done in one of two ways:

- CoExSiST calibrates **microscopically**: in a Discrete Element Method (DEM) context, all simulated particles follow their experimental counterparts *exactly*. Naturally, this technique is limited to dilute systems and experimental imaging techniques that can capture the 3D position of *all* moving particles (e.g. PIV) - however, it provides information about the fundamental aspects of particle collision.
- ACCES calibrates / optimises **macroscopically**: a given simulation reproduces a system-specific *macroscopic* quantity (e.g. residence time distribution, angle of repose). This technique is completely agnostic to the simulation method and the quantity to be reproduced. For example, it can train *coarse-grained DEM* simulations, using larger meso-particles to model multiple smaller ones.


ACCES is ready for production use; it was successfully used to calibrate coarse-grained DEM digital twins of [GranuTools](https://www.granutools.com/en/) equipment (Andrei Leonard Nicusan and Dominik Werner, *paper under review*), CFDEM fluidised beds (Hanqiao Cha, *paper under review*) and even signal processing parameters in a PET scanner model (Matthew Herald, *paper under review*).


![Calibrated GranuDrum](/docs/source/_static/calibrated.png?raw=true "Calibrated GranuDrum.")
*Example of an ACCES-calibrated DEM Digital Twin of a GranuTools GranuDrum; the calibration was done effectively against a single experimental data point - a photograph of the free surface shape yielded by MCC particles (left panel). The occupancy grid of a LIGGGHTS simulation was optimised against the free surface shape (right panel). The two superimposed grids amount to 4 mm² dissimilarity (dark blue pixels, middle panel).*


ACCES was implemented in the `coexist.Access` class, providing an interface that is easy to use, but powerful enough to **automatically parallelise arbitrary Python scripts** through code inspection and metaprogramming. It was used successfully from laptop-scale shared-memory machines to multi-node supercomputing clusters.





## Getting Started

This is a pure Python package that does not require any extra system configuration, supporting Python 3.6 and above (though it might work with even older versions) - to install it from PyPI, simply run:

```
pip install coexist
```

Or you can install the development version from the GitHub repository:

```
pip install git+https://github.com/uob-positron-imaging-centre/Coexist
```




### Examples

The [documentation](https://coexist.readthedocs.io/) website contains an ACCES [tutorial](https://coexist.readthedocs.io/en/latest/tutorials/index.html) with explained code and output figures produced by `coexist`; all public functionality is fully documented in the [manual](https://coexist.readthedocs.io/en/latest/manual/index.html).

Want something more hands on? Check out the `examples` folder for example scripts using `coexist.Coexist` and `coexist.Access`; `examples/access_simple` is a very simple, hackable example script (remember that the `simulation_script.py` can execute *anything*). For a more involved, complete calibration of a GranuTools GranuDrum digital twin - using LIGGGHTS - see our collection of peer-reviewed digital twins [repository](https://github.com/uob-positron-imaging-centre/DigitalTwins).


![GranuDrum ACCES Example](/docs/source/_static/access_example.png?raw=true "GranuDrum ACCES Example.")


The `coexist.plots` submodule can plot the convergence of ACCES onto optimally-calibrated parameters - or check intermediate results while your 50-hour simulations runs:


![Convergence Plot](/docs/source/_static/access_convergence.png?raw=true "ACCES Convergence Plot.")





### Show me some code already...

Alright, here's the gist of it: instead of having to rewrite your complex simulation into a function for calibration / optimisation (which is what virtually all optimisation frameworks require...), ACCES takes in an _entire simulation script_; here's a simple example, say in a file called `simulation_script.py`:

```python
# File: simulation_script.py

# ACCESS PARAMETERS START
import coexist

parameters = coexist.create_parameters(
    variables = ["CED", "CoR", "Epsilon", "Mu"],
    minimums = [-5, -5, -5, -5],
    maximums = [+5, +5, +5, +5],
    values = [0, 0, 0, 0],          # Optional, initial guess
)

access_id = 0                       # Optional, unique ID for each simulation
# ACCESS PARAMETERS END


# Extract the free parameters' values - ACCES will modify / optimise them.
x, y, z, t = parameters["value"]


# Define the error value in *any* way - run a simulation, analyse data, etc.
#
# For simplicity, here's an analytical 4D Himmelblau function with 8 global
# and 2 local minima - the initial guess is very close to the local one!
error = (x**2 + y - 11)**2 + (x + y**2 - 7)**2 + z * t
```

Then you can run in a separate script or Python console:

```python
# File: access_learn.py

import coexist

# Use ACCESS to learn a simulation's parameters
access = coexist.Access("simulation_script.py")
access.learn(
    num_solutions = 10,         # Number of solutions per epoch
    target_sigma = 0.1,         # Target std-dev (accuracy) of solution
    random_seed = 42,           # Reproducible / deterministic optimisation
)

# Or simply
# coexist.Access("simulation_script.py").learn(random_seed = 42)
```

That's it - ACCES will take the `simulation_script.py`, modify the free `parameters`, run the script in parallel (either on all processors of your local machine or on a distributed supercomputer) and optimise the `error` you defined; it'll look something like this:

```
================================================================================
Starting ACCES run at 08:32:23 on 01/02/2022 in directory `access_seed42`.
================================================================================
Epoch    0 | Population 10
--------------------------------------------------------------------------------
Scaled overall standard deviation: 1.0
             CED       CoR   Epsilon        Mu
estimate     0.0  0.000000  0.000000  0.000000
uncertainty  4.0  4.000050  4.000100  4.000150
scaled_std   1.0  1.000013  1.000025  1.000038

        CED       CoR   Epsilon        Mu       error
0  1.218868 -4.159988  3.001880  3.762400  331.093228
1 -3.095859 -4.967675  0.511374 -1.265018  252.732762
2 -0.067205 -3.412218  3.517680  3.111284  239.466423
3  0.264123  4.509021  1.870084 -3.437299  219.638764
4  1.475003 -3.835578  3.513889 -0.199711  243.967225
5 -0.739449 -2.723752  4.825957 -0.618141  170.752129
6 -1.713311 -1.408552  2.129290  1.461831  138.135979
7  1.650930  1.723306  2.333195 -1.625721   44.785098
8 -2.048971 -3.255132  2.463979  4.516059  114.660635
9 -0.455790 -3.360668 -3.298007  2.602469  206.454823
Total function evaluations: 10

================================================================================
Epoch    1 | Population 10
--------------------------------------------------------------------------------
...
<output truncated>
...
================================================================================
Epoch   31 | Population 10
--------------------------------------------------------------------------------
Scaled overall standard deviation: 0.1032642976669169
                  CED       CoR   Epsilon        Mu
estimate     3.621098 -1.748473  4.999445 -4.992881
uncertainty  0.052355  0.078080  0.284105  0.180196
scaled_std   0.013089  0.019520  0.071026  0.045049

          CED       CoR   Epsilon        Mu      error
310  3.574473 -1.650134  4.999374 -4.901913 -23.996813
311  3.582026 -1.756384  4.863496 -4.973417 -24.071692
312  3.628789 -1.616891  4.859678 -4.992656 -23.386000
313  3.662351 -1.782704  4.982040 -4.998059 -24.478008
314  3.594971 -1.725715  4.999877 -4.898257 -24.269162
315  3.577725 -1.744971  4.998411 -4.956502 -24.629198
316  3.613914 -1.747412  4.983253 -4.996630 -24.690880
317  3.579212 -1.774811  4.852262 -4.972910 -24.055219
318  3.634952 -1.784927  4.999971 -4.999863 -24.783959
319  3.647169 -1.872419  4.999971 -4.978640 -24.685207
Total function evaluations: 320


Optimal solution found within `target_sigma`, i.e. 10.0%:
  sigma = 0.08390460663313491 < 0.1

================================================================================
The best result was found in 32 epochs:
              CED       CoR   Epsilon        Mu      error
  value  3.569249 -1.813354  4.995112 -4.994052 -24.920092
  std    0.042168  0.060116  0.203900  0.140874

These results were found for the job:
  access_seed42/results/parameters.284.pickle
================================================================================
```

And you can access (pun intended) the results - even as ACCES is running - using:

```python
>>> import coexist
>>> data = coexist.AccessData.read("access_seed42")
>>> data
AccessData
--------------------------------------------------------------------------------
paths          ╎ AccessPaths(...)
parameters     ╎             value  min  max     sigma
               ╎ CED      3.569249 -5.0  5.0  0.052355
               ╎ CoR     -1.813354 -5.0  5.0  0.078080
               ╎ Epsilon  4.995112 -5.0  5.0  0.284105
               ╎ Mu      -4.994052 -5.0  5.0  0.180196
population     ╎ 10
num_epochs     ╎ 32
target         ╎ 0.1
seed           ╎ 42
epochs         ╎ DataFrame(CED_mean, CoR_mean, Epsilon_mean, Mu_mean, CED_std,
               ╎           CoR_std, Epsilon_std, Mu_std, overall_std)
epochs_scaled  ╎ DataFrame(CED_mean, CoR_mean, Epsilon_mean, Mu_mean, CED_std,
               ╎           CoR_std, Epsilon_std, Mu_std, overall_std)
results        ╎ DataFrame(CED, CoR, Epsilon, Mu, error)
results_scaled ╎ DataFrame(CED, CoR, Epsilon, Mu, error)
```

In this case a global optimum was found within 320 evaluations - this is of course problem-dependent, but you'll see that the optimum is often found much earlier if you check intermediate results (which you probably will when calibrating / optimising long-running simulations).

A tutorial with more detailed explanations is available [here](https://coexist.readthedocs.io/en/latest/tutorials/index.html), including generating plots, checking intermediate results and running simulations on a SLURM-managed distributed cluster.




## Contributing

This library aims to be the state-of-the-art for simulation calibration, developed in the open using modern, collaborative coding approaches - no dragons shall be dwelling in the codebase. You are more than welcome to contribute to this library in the form of code improvements, documentation or helpful examples; please submit them either as:

- GitHub issues.
- Pull requests.
- Email me at <a.l.nicusan@gmail.com>.

We are more than happy to discuss the library architecture and calibration / optimisation approach with any potential contributors and users.




## Acknowledgements and Funding

The authors gratefully acknowledge funding from the following UK funding bodies and industrial partners:

**M²E³D: Multiphase Materials Exploration via Evolutionary Equation Discovery**  
Royce Materials 4.0 Feasibility and Pilot Scheme Grant, £57,477  

**CoExSiST: Coupled Experimental-Simulational Technique**  
EPSRC MAPP Grant, Feasibility Study, £60,246  

**ACCES: Autonomous Calibration and Characterisation via Evolutionary Simulation**  
EPSRC IAA, Follow-Up Grant to CoExSiST, £52,762  

**Improving ACCES: Towards the Multi-Tool Multi-Parameter Optimisation of Complex Particulate Systems**  
EPSRC MAPP, Grant, £48,871 + £48,871 matched funding from GranuTools Belgium  

Thank you.




## Citing

If you use this library in your research, you are kindly asked to cite:

> [Paper after publication]


Until the ACCES paper is published, you may cite this repository:

> Nicusan AL, Werner D, Sykes J, Seville JPK, Windows-Yule CR. ACCES: Autonomous Characterisation and Calibration via Evolutionary Simulation. GitHub repository. 2022 February 1.


ACCES is built on top of the excellent CMA-ES evolutionary algorithm - specifically the [`pycma`](https://github.com/CMA-ES/pycma) implementation. If you use ACCES in your research, please also cite:

> Nikolaus Hansen, Youhei Akimoto, and Petr Baudis. CMA-ES/pycma on Github. Zenodo, DOI:10.5281/zenodo.2559634, February 2019.




## License and Commercial Integration

This library - in its general, domain-agnostic form - is free and open-source, published under the GNU General Public License v3.0.

If you are a company and would like to integrate ACCESS into your work - e.g. ACCESS-enabled equipment or general simulation calibration - please send an email to `a.l.nicusan@bham.ac.uk` to discuss commercial development of specific tools for your application. Relicensing for a closed-source / commercial project can be considered on an individual basis.


Copyright (C) 2020-2023 the Coexist developers. Until now, this library was built directly or indirectly through the brain-time of:

- Andrei Leonard Nicusan (University of Birmingham)
- Dominik Werner (University of Birmingham)
- Jack Sykes (University of Birmingham)
- Dr. Kit Windows-Yule (University of Birmingham)
- Prof. Jonathan Seville (University of Birmingham)
- Albert Bauer (TU Berlin)

Thank you.

            

Raw data

            {
    "_id": null,
    "home_page": "https://github.com/uob-positron-imaging-centre/Coexist",
    "name": "coexist",
    "maintainer": "",
    "docs_url": null,
    "requires_python": ">=3.6",
    "maintainer_email": "",
    "keywords": "simulation optimization calibration parameter-estimation parameter-tuning physics-simulation computational-fluid-dynamics granular DEM discrete-element-method medical-imaging",
    "author": "Andrei Leonard Nicusan <a.l.nicusan@bham.ac.uk>, Dominik Werner <d.werner.1@bham.ac.uk>, Jack Sykes <jas653@student.bham.ac.uk>",
    "author_email": "",
    "download_url": "https://files.pythonhosted.org/packages/b7/c8/f2b192f0729cef9f96686069638bfb857dcb431eceeeecc9d668f6b65f90/coexist-0.3.2.tar.gz",
    "platform": null,
    "description": "[![PyPI version](https://badge.fury.io/py/coexist.svg)](https://badge.fury.io/py/coexist)\n[![Documentation Status](https://readthedocs.org/projects/coexist/badge/?version=latest)](https://coexist.readthedocs.io/en/latest/?badge=latest)\n[![CI Status](https://github.com/uob-positron-imaging-centre/Coexist/actions/workflows/ci.yml/badge.svg)](https://github.com/uob-positron-imaging-centre/Coexist/actions/workflows/ci.yml)\n[![Language grade: Python](https://img.shields.io/lgtm/grade/python/g/uob-positron-imaging-centre/Coexist.svg?logo=lgtm&logoWidth=18)](https://lgtm.com/projects/g/uob-positron-imaging-centre/Coexist/context:python)\n[![Colab example](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1p7OwnaYgENwK4DTn_6QETX4ajwVFeza0?usp=sharing)\n[![License: GPL-3.0](https://img.shields.io/github/license/uob-positron-imaging-centre/Coexist?style=flat-square)](https://github.com/uob-positron-imaging-centre/Coexist)\n\n\n# CoExSiST & ACCES\n### Data-Driven Evolutionary Calibration & Optimisation of Simulations\n\nA Python library for autonomously learning simulation parameters from experimental data, from the *micro* to the *macro*, from laptops to clusters. This is done using either of two closely related frameworks:\n\n- **CoExSiST**: Coupled Experimental-Simulational Study Tool.\n- **ACCES**: Autonomous Characterisation and Calibration via Evolutionary Simulation. \n\nBoth libraries learn a given set of free parameters, such that an experiment is synchronised with an equivalent simulation; this synchronisation is done in one of two ways:\n\n- CoExSiST calibrates **microscopically**: in a Discrete Element Method (DEM) context, all simulated particles follow their experimental counterparts *exactly*. Naturally, this technique is limited to dilute systems and experimental imaging techniques that can capture the 3D position of *all* moving particles (e.g. PIV) - however, it provides information about the fundamental aspects of particle collision.\n- ACCES calibrates / optimises **macroscopically**: a given simulation reproduces a system-specific *macroscopic* quantity (e.g. residence time distribution, angle of repose). This technique is completely agnostic to the simulation method and the quantity to be reproduced. For example, it can train *coarse-grained DEM* simulations, using larger meso-particles to model multiple smaller ones.\n\n\nACCES is ready for production use; it was successfully used to calibrate coarse-grained DEM digital twins of [GranuTools](https://www.granutools.com/en/) equipment (Andrei Leonard Nicusan and Dominik Werner, *paper under review*), CFDEM fluidised beds (Hanqiao Cha, *paper under review*) and even signal processing parameters in a PET scanner model (Matthew Herald, *paper under review*).\n\n\n![Calibrated GranuDrum](/docs/source/_static/calibrated.png?raw=true \"Calibrated GranuDrum.\")\n*Example of an ACCES-calibrated DEM Digital Twin of a GranuTools GranuDrum; the calibration was done effectively against a single experimental data point - a photograph of the free surface shape yielded by MCC particles (left panel). The occupancy grid of a LIGGGHTS simulation was optimised against the free surface shape (right panel). The two superimposed grids amount to 4 mm\u00b2 dissimilarity (dark blue pixels, middle panel).*\n\n\nACCES was implemented in the `coexist.Access` class, providing an interface that is easy to use, but powerful enough to **automatically parallelise arbitrary Python scripts** through code inspection and metaprogramming. It was used successfully from laptop-scale shared-memory machines to multi-node supercomputing clusters.\n\n\n\n\n\n## Getting Started\n\nThis is a pure Python package that does not require any extra system configuration, supporting Python 3.6 and above (though it might work with even older versions) - to install it from PyPI, simply run:\n\n```\npip install coexist\n```\n\nOr you can install the development version from the GitHub repository:\n\n```\npip install git+https://github.com/uob-positron-imaging-centre/Coexist\n```\n\n\n\n\n### Examples\n\nThe [documentation](https://coexist.readthedocs.io/) website contains an ACCES [tutorial](https://coexist.readthedocs.io/en/latest/tutorials/index.html) with explained code and output figures produced by `coexist`; all public functionality is fully documented in the [manual](https://coexist.readthedocs.io/en/latest/manual/index.html).\n\nWant something more hands on? Check out the `examples` folder for example scripts using `coexist.Coexist` and `coexist.Access`; `examples/access_simple` is a very simple, hackable example script (remember that the `simulation_script.py` can execute *anything*). For a more involved, complete calibration of a GranuTools GranuDrum digital twin - using LIGGGHTS - see our collection of peer-reviewed digital twins [repository](https://github.com/uob-positron-imaging-centre/DigitalTwins).\n\n\n![GranuDrum ACCES Example](/docs/source/_static/access_example.png?raw=true \"GranuDrum ACCES Example.\")\n\n\nThe `coexist.plots` submodule can plot the convergence of ACCES onto optimally-calibrated parameters - or check intermediate results while your 50-hour simulations runs:\n\n\n![Convergence Plot](/docs/source/_static/access_convergence.png?raw=true \"ACCES Convergence Plot.\")\n\n\n\n\n\n### Show me some code already...\n\nAlright, here's the gist of it: instead of having to rewrite your complex simulation into a function for calibration / optimisation (which is what virtually all optimisation frameworks require...), ACCES takes in an _entire simulation script_; here's a simple example, say in a file called `simulation_script.py`:\n\n```python\n# File: simulation_script.py\n\n# ACCESS PARAMETERS START\nimport coexist\n\nparameters = coexist.create_parameters(\n    variables = [\"CED\", \"CoR\", \"Epsilon\", \"Mu\"],\n    minimums = [-5, -5, -5, -5],\n    maximums = [+5, +5, +5, +5],\n    values = [0, 0, 0, 0],          # Optional, initial guess\n)\n\naccess_id = 0                       # Optional, unique ID for each simulation\n# ACCESS PARAMETERS END\n\n\n# Extract the free parameters' values - ACCES will modify / optimise them.\nx, y, z, t = parameters[\"value\"]\n\n\n# Define the error value in *any* way - run a simulation, analyse data, etc.\n#\n# For simplicity, here's an analytical 4D Himmelblau function with 8 global\n# and 2 local minima - the initial guess is very close to the local one!\nerror = (x**2 + y - 11)**2 + (x + y**2 - 7)**2 + z * t\n```\n\nThen you can run in a separate script or Python console:\n\n```python\n# File: access_learn.py\n\nimport coexist\n\n# Use ACCESS to learn a simulation's parameters\naccess = coexist.Access(\"simulation_script.py\")\naccess.learn(\n    num_solutions = 10,         # Number of solutions per epoch\n    target_sigma = 0.1,         # Target std-dev (accuracy) of solution\n    random_seed = 42,           # Reproducible / deterministic optimisation\n)\n\n# Or simply\n# coexist.Access(\"simulation_script.py\").learn(random_seed = 42)\n```\n\nThat's it - ACCES will take the `simulation_script.py`, modify the free `parameters`, run the script in parallel (either on all processors of your local machine or on a distributed supercomputer) and optimise the `error` you defined; it'll look something like this:\n\n```\n================================================================================\nStarting ACCES run at 08:32:23 on 01/02/2022 in directory `access_seed42`.\n================================================================================\nEpoch    0 | Population 10\n--------------------------------------------------------------------------------\nScaled overall standard deviation: 1.0\n             CED       CoR   Epsilon        Mu\nestimate     0.0  0.000000  0.000000  0.000000\nuncertainty  4.0  4.000050  4.000100  4.000150\nscaled_std   1.0  1.000013  1.000025  1.000038\n\n        CED       CoR   Epsilon        Mu       error\n0  1.218868 -4.159988  3.001880  3.762400  331.093228\n1 -3.095859 -4.967675  0.511374 -1.265018  252.732762\n2 -0.067205 -3.412218  3.517680  3.111284  239.466423\n3  0.264123  4.509021  1.870084 -3.437299  219.638764\n4  1.475003 -3.835578  3.513889 -0.199711  243.967225\n5 -0.739449 -2.723752  4.825957 -0.618141  170.752129\n6 -1.713311 -1.408552  2.129290  1.461831  138.135979\n7  1.650930  1.723306  2.333195 -1.625721   44.785098\n8 -2.048971 -3.255132  2.463979  4.516059  114.660635\n9 -0.455790 -3.360668 -3.298007  2.602469  206.454823\nTotal function evaluations: 10\n\n================================================================================\nEpoch    1 | Population 10\n--------------------------------------------------------------------------------\n...\n<output truncated>\n...\n================================================================================\nEpoch   31 | Population 10\n--------------------------------------------------------------------------------\nScaled overall standard deviation: 0.1032642976669169\n                  CED       CoR   Epsilon        Mu\nestimate     3.621098 -1.748473  4.999445 -4.992881\nuncertainty  0.052355  0.078080  0.284105  0.180196\nscaled_std   0.013089  0.019520  0.071026  0.045049\n\n          CED       CoR   Epsilon        Mu      error\n310  3.574473 -1.650134  4.999374 -4.901913 -23.996813\n311  3.582026 -1.756384  4.863496 -4.973417 -24.071692\n312  3.628789 -1.616891  4.859678 -4.992656 -23.386000\n313  3.662351 -1.782704  4.982040 -4.998059 -24.478008\n314  3.594971 -1.725715  4.999877 -4.898257 -24.269162\n315  3.577725 -1.744971  4.998411 -4.956502 -24.629198\n316  3.613914 -1.747412  4.983253 -4.996630 -24.690880\n317  3.579212 -1.774811  4.852262 -4.972910 -24.055219\n318  3.634952 -1.784927  4.999971 -4.999863 -24.783959\n319  3.647169 -1.872419  4.999971 -4.978640 -24.685207\nTotal function evaluations: 320\n\n\nOptimal solution found within `target_sigma`, i.e. 10.0%:\n  sigma = 0.08390460663313491 < 0.1\n\n================================================================================\nThe best result was found in 32 epochs:\n              CED       CoR   Epsilon        Mu      error\n  value  3.569249 -1.813354  4.995112 -4.994052 -24.920092\n  std    0.042168  0.060116  0.203900  0.140874\n\nThese results were found for the job:\n  access_seed42/results/parameters.284.pickle\n================================================================================\n```\n\nAnd you can access (pun intended) the results - even as ACCES is running - using:\n\n```python\n>>> import coexist\n>>> data = coexist.AccessData.read(\"access_seed42\")\n>>> data\nAccessData\n--------------------------------------------------------------------------------\npaths          \u254e AccessPaths(...)\nparameters     \u254e             value  min  max     sigma\n               \u254e CED      3.569249 -5.0  5.0  0.052355\n               \u254e CoR     -1.813354 -5.0  5.0  0.078080\n               \u254e Epsilon  4.995112 -5.0  5.0  0.284105\n               \u254e Mu      -4.994052 -5.0  5.0  0.180196\npopulation     \u254e 10\nnum_epochs     \u254e 32\ntarget         \u254e 0.1\nseed           \u254e 42\nepochs         \u254e DataFrame(CED_mean, CoR_mean, Epsilon_mean, Mu_mean, CED_std,\n               \u254e           CoR_std, Epsilon_std, Mu_std, overall_std)\nepochs_scaled  \u254e DataFrame(CED_mean, CoR_mean, Epsilon_mean, Mu_mean, CED_std,\n               \u254e           CoR_std, Epsilon_std, Mu_std, overall_std)\nresults        \u254e DataFrame(CED, CoR, Epsilon, Mu, error)\nresults_scaled \u254e DataFrame(CED, CoR, Epsilon, Mu, error)\n```\n\nIn this case a global optimum was found within 320 evaluations - this is of course problem-dependent, but you'll see that the optimum is often found much earlier if you check intermediate results (which you probably will when calibrating / optimising long-running simulations).\n\nA tutorial with more detailed explanations is available [here](https://coexist.readthedocs.io/en/latest/tutorials/index.html), including generating plots, checking intermediate results and running simulations on a SLURM-managed distributed cluster.\n\n\n\n\n## Contributing\n\nThis library aims to be the state-of-the-art for simulation calibration, developed in the open using modern, collaborative coding approaches - no dragons shall be dwelling in the codebase. You are more than welcome to contribute to this library in the form of code improvements, documentation or helpful examples; please submit them either as:\n\n- GitHub issues.\n- Pull requests.\n- Email me at <a.l.nicusan@gmail.com>.\n\nWe are more than happy to discuss the library architecture and calibration / optimisation approach with any potential contributors and users.\n\n\n\n\n## Acknowledgements and Funding\n\nThe authors gratefully acknowledge funding from the following UK funding bodies and industrial partners:\n\n**M\u00b2E\u00b3D: Multiphase Materials Exploration via Evolutionary Equation Discovery**  \nRoyce Materials 4.0 Feasibility and Pilot Scheme Grant, \u00a357,477  \n\n**CoExSiST: Coupled Experimental-Simulational Technique**  \nEPSRC MAPP Grant, Feasibility Study, \u00a360,246  \n\n**ACCES: Autonomous Calibration and Characterisation via Evolutionary Simulation**  \nEPSRC IAA, Follow-Up Grant to CoExSiST, \u00a352,762  \n\n**Improving ACCES: Towards the Multi-Tool Multi-Parameter Optimisation of Complex Particulate Systems**  \nEPSRC MAPP, Grant, \u00a348,871 + \u00a348,871 matched funding from GranuTools Belgium  \n\nThank you.\n\n\n\n\n## Citing\n\nIf you use this library in your research, you are kindly asked to cite:\n\n> [Paper after publication]\n\n\nUntil the ACCES paper is published, you may cite this repository:\n\n> Nicusan AL, Werner D, Sykes J, Seville JPK, Windows-Yule CR. ACCES: Autonomous Characterisation and Calibration via Evolutionary Simulation. GitHub repository. 2022 February 1.\n\n\nACCES is built on top of the excellent CMA-ES evolutionary algorithm - specifically the [`pycma`](https://github.com/CMA-ES/pycma) implementation. If you use ACCES in your research, please also cite:\n\n> Nikolaus Hansen, Youhei Akimoto, and Petr Baudis. CMA-ES/pycma on Github. Zenodo, DOI:10.5281/zenodo.2559634, February 2019.\n\n\n\n\n## License and Commercial Integration\n\nThis library - in its general, domain-agnostic form - is free and open-source, published under the GNU General Public License v3.0.\n\nIf you are a company and would like to integrate ACCESS into your work - e.g. ACCESS-enabled equipment or general simulation calibration - please send an email to `a.l.nicusan@bham.ac.uk` to discuss commercial development of specific tools for your application. Relicensing for a closed-source / commercial project can be considered on an individual basis.\n\n\nCopyright (C) 2020-2023 the Coexist developers. Until now, this library was built directly or indirectly through the brain-time of:\n\n- Andrei Leonard Nicusan (University of Birmingham)\n- Dominik Werner (University of Birmingham)\n- Jack Sykes (University of Birmingham)\n- Dr. Kit Windows-Yule (University of Birmingham)\n- Prof. Jonathan Seville (University of Birmingham)\n- Albert Bauer (TU Berlin)\n\nThank you.\n",
    "bugtrack_url": null,
    "license": "GNU",
    "summary": "Learning simulation parameters from experimental data, from the micro to the macro, from the laptop to the cluster.",
    "version": "0.3.2",
    "project_urls": {
        "Homepage": "https://github.com/uob-positron-imaging-centre/Coexist"
    },
    "split_keywords": [
        "simulation",
        "optimization",
        "calibration",
        "parameter-estimation",
        "parameter-tuning",
        "physics-simulation",
        "computational-fluid-dynamics",
        "granular",
        "dem",
        "discrete-element-method",
        "medical-imaging"
    ],
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "blake2b_256": "8146bdb21d6ccf75666af516cf7020f7dfcdf91785e75b1f5ee107e221401c45",
                "md5": "d779dfcfc9d0cda537c1ed4f1523c928",
                "sha256": "47b09e63847220bd4a577edf9eb725cadf0c089ef535219ed7ea2bae17001c61"
            },
            "downloads": -1,
            "filename": "coexist-0.3.2-py2.py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "d779dfcfc9d0cda537c1ed4f1523c928",
            "packagetype": "bdist_wheel",
            "python_version": "py2.py3",
            "requires_python": ">=3.6",
            "size": 82987,
            "upload_time": "2023-06-11T02:52:55",
            "upload_time_iso_8601": "2023-06-11T02:52:55.420236Z",
            "url": "https://files.pythonhosted.org/packages/81/46/bdb21d6ccf75666af516cf7020f7dfcdf91785e75b1f5ee107e221401c45/coexist-0.3.2-py2.py3-none-any.whl",
            "yanked": false,
            "yanked_reason": null
        },
        {
            "comment_text": "",
            "digests": {
                "blake2b_256": "b7c8f2b192f0729cef9f96686069638bfb857dcb431eceeeecc9d668f6b65f90",
                "md5": "d9175cc580f5c141e0c6faf1a7afd821",
                "sha256": "c82d659de5d9a2ebeea7490166f44dd1d3e5788dcd831c0940a04b9eb124dd36"
            },
            "downloads": -1,
            "filename": "coexist-0.3.2.tar.gz",
            "has_sig": false,
            "md5_digest": "d9175cc580f5c141e0c6faf1a7afd821",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": ">=3.6",
            "size": 86801,
            "upload_time": "2023-06-11T02:52:57",
            "upload_time_iso_8601": "2023-06-11T02:52:57.464611Z",
            "url": "https://files.pythonhosted.org/packages/b7/c8/f2b192f0729cef9f96686069638bfb857dcb431eceeeecc9d668f6b65f90/coexist-0.3.2.tar.gz",
            "yanked": false,
            "yanked_reason": null
        }
    ],
    "upload_time": "2023-06-11 02:52:57",
    "github": true,
    "gitlab": false,
    "bitbucket": false,
    "codeberg": false,
    "github_user": "uob-positron-imaging-centre",
    "github_project": "Coexist",
    "travis_ci": false,
    "coveralls": false,
    "github_actions": true,
    "requirements": [],
    "lcname": "coexist"
}
        
Elapsed time: 0.09511s