https://github.com/pgdr/mcelfish
https://github.com/pgdr/mcelfish
Last synced: about 1 year ago
JSON representation
- Host: GitHub
- URL: https://github.com/pgdr/mcelfish
- Owner: pgdr
- License: mit
- Created: 2020-10-07T13:06:52.000Z (over 5 years ago)
- Default Branch: main
- Last Pushed: 2020-12-01T15:17:47.000Z (over 5 years ago)
- Last Synced: 2025-02-01T23:42:08.508Z (over 1 year ago)
- Language: Python
- Size: 169 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# mcelfish -- MCMC Removal Sampling
Do _removal sampling_ with MCMC, Zippin, or Carle-Strub.
**Usage:**
```
mcelfish --data 19 17 13 1 1
```
```
mcelfish --plot --data 19 17 13 1 1
```
## Installing
```
pip install mcelfish
```
or from source
```
git clone git@github.com:pgdr/mcelfish
pip install -e ./mcelfish
```
## Removal sampling
Suppose that you want to
[estimate the number of apples in an apple tree](https://stats.stackexchange.com/questions/491165/estimating-the-number-of-apples-in-an-apple-tree-using-mcmc)
by repeatedly kicking the tree and counting how many apples fall down.
We make an assumption that there is a (constant) _p_ between 0 and 1,
such that the probability that any given apple falls down when you kick
the tree is _p_.
Suppose that the apple tree contains _N_ apples. Given a series of
kicks, we want to estimate _N_ and _p_.
We call this problem _removal sampling_.
### Examples
Suppose that `[100, 10, 1, 0]` apples fall down. In this case, we may
estimate _p_ to be approximately 0.9, and _N_ approximately 111.
Suppose that `[19, 17, 13, 1, 1]` apples fall down. In this more
complicated case, Bayesian statistics tell us that we should expect
_53 ≤ N ≤ 57_ and _0.4 ≤ p ≤ 0.45_.
## Advanced MCMC usage
This package uses [pymc3](https://pypi.org/project/pymc3/). Tuning the
MCMC run can be done with the parameters `--samplings` and `--tunings`,
e.g.
```
mcelfish --samples 10000 --tunings 5000 --plot --data 19 17 13 1 1
```
We can use `--beta` to output the beta parameters for the posterior.