https://github.com/johannesbuchner/stagedstan
Staged Stan Python package
https://github.com/johannesbuchner/stagedstan
Last synced: 7 months ago
JSON representation
Staged Stan Python package
- Host: GitHub
- URL: https://github.com/johannesbuchner/stagedstan
- Owner: JohannesBuchner
- License: other
- Created: 2014-02-03T00:22:54.000Z (over 11 years ago)
- Default Branch: master
- Last Pushed: 2014-02-03T00:59:24.000Z (over 11 years ago)
- Last Synced: 2025-01-23T20:51:32.467Z (8 months ago)
- Language: Python
- Size: 148 KB
- Stars: 1
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.rst
- License: LICENSE
Awesome Lists containing this project
README
StagedStan Python package
===========================
This package runs the Stan_ Markov Chain Monte Carlo sampler.
* StagedStan works similar to "make", namely in a deterministic fashion.
* Runs consist of *compilation*, *optimization* and *sampling* (starting from the best fit).
* When aborted, it continues where it was stopped.
* Unlike PyStan_, all results are stored in a file structure.
* Each output is uniquely associated to the Stan code, data and seed used. If neither of them change, the result is not re-computed.
* This also allows you to execute Stan on another machine, and to copy the output over.
* StagedStan can parse the summary files of Stan runs, and write data dump files.
* StagedStan allows the input of data from Python (e.g. numpy arrays).
Installation
-------------
Download and install the python package as usual (python setup.py).
StagedStan requires Stan. Let the STANDIR environment variable point to the Stan directory (where makefile lives).
.. _Stan: http://mc-stan.org/
.. _PyStan: http://mc-stan.org/pystan.html
Example
---------
Python code::
stancode = """
data {
int N;
real stdev;
real obs[N,2];
}
parameters {
real mean1;
real mean2;
}
model {
for (i in 1:N) {
obs[i,1] ~ normal(mean1, stdev);
obs[i,2] ~ normal(mean2, stdev);
}
}
"""
numpy.random.seed(0)
obs = numpy.transpose([
numpy.random.normal(2, 0.1, size=10),
numpy.random.normal(-1, 0.1, size=10)
])
data = dict(N=10, stdev = 0.1, obs = obs)
print 'Best fit:'
for k,v in get_best_fit(stancode, data).iteritems():
if '__' not in k:
print ' %-20s : %s ' % (k, v)
print
print 'Samples:'
for k,v in get_samples(stancode, data, shuffle=True).iteritems():
if '__' not in k:
print ' %-20s : %s ' % (k, v)
print
Output (parts are omitted)::
[...] building executable [...]
Running code: ('./code-f2808a03aef7f025449bb9df437764b6318acbb3/code.exe', 'output', 'file=./code-f2808a03aef7f025449bb9df437764b6318acbb3/data-2fbafe7c1993572a57131ef2a800f4072a088f3b/bestfit1.out', 'data', 'file=./code-f2808a03aef7f025449bb9df437764b6318acbb3/data-2fbafe7c1993572a57131ef2a800f4072a088f3b/data.R', 'random', 'seed=1', 'optimize')
Best fit:
mean1 : 2.07380231707
mean2 : -0.959935398484
Running code: ('./code-f2808a03aef7f025449bb9df437764b6318acbb3/code.exe', 'output', 'file=./code-f2808a03aef7f025449bb9df437764b6318acbb3/data-2fbafe7c1993572a57131ef2a800f4072a088f3b/samples1.out', 'data', 'file=./code-f2808a03aef7f025449bb9df437764b6318acbb3/data-2fbafe7c1993572a57131ef2a800f4072a088f3b/data.R', 'init=./code-f2808a03aef7f025449bb9df437764b6318acbb3/data-2fbafe7c1993572a57131ef2a800f4072a088f3b/bestfit1.R', 'random', 'seed=1', 'sample')
Gradient evaluation took 0 seconds
1000 transitions using 10 leapfrog steps per transition would take 0 seconds.
Adjust your expectations accordingly!
Iteration: 1 / 2000 [ 0%] (Warmup)
Iteration: 100 / 2000 [ 5%] (Warmup)
Iteration: 200 / 2000 [ 10%] (Warmup)
Iteration: 300 / 2000 [ 15%] (Warmup)
Iteration: 400 / 2000 [ 20%] (Warmup)
Iteration: 500 / 2000 [ 25%] (Warmup)
Iteration: 600 / 2000 [ 30%] (Warmup)
Iteration: 700 / 2000 [ 35%] (Warmup)
Iteration: 800 / 2000 [ 40%] (Warmup)
Iteration: 900 / 2000 [ 45%] (Warmup)
Iteration: 1000 / 2000 [ 50%] (Warmup)
Iteration: 1001 / 2000 [ 50%] (Sampling)
Iteration: 1100 / 2000 [ 55%] (Sampling)
Iteration: 1200 / 2000 [ 60%] (Sampling)
Iteration: 1300 / 2000 [ 65%] (Sampling)
Iteration: 1400 / 2000 [ 70%] (Sampling)
Iteration: 1500 / 2000 [ 75%] (Sampling)
Iteration: 1600 / 2000 [ 80%] (Sampling)
Iteration: 1700 / 2000 [ 85%] (Sampling)
Iteration: 1800 / 2000 [ 90%] (Sampling)
Iteration: 1900 / 2000 [ 95%] (Sampling)
Iteration: 2000 / 2000 [100%] (Sampling)
Elapsed Time: 0.01 seconds (Warm-up)
0.02 seconds (Sampling)
0.03 seconds (Total)
Samples:
mean1 : [ 2.10492 2.04165 2.11458 2.04987 2.06028 2.05426 2.09401 ...]
mean2 : [-0.897088 -0.946299 -0.894294 -0.963678 -1.00048 -0.983169 -0.9965 ...]
The following files are created:
* test/code-f2808a/code.exe -- compiled executable
* test/code-f2808a/data-2fbaf/bestfit1.out -- results of the optimization
* test/code-f2808a/data-2fbaf/samples1.out -- results of the sampling
Read `the code `_ (only ~160 lines) for API documentation.