Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/filipspl/bayesian-rf-knime-scikit
Bayesian optimization of RF via scikit in KNIME
https://github.com/filipspl/bayesian-rf-knime-scikit
bayesian-optimization knime knime-analytics-platform python random-forest scikit-learn
Last synced: 25 days ago
JSON representation
Bayesian optimization of RF via scikit in KNIME
- Host: GitHub
- URL: https://github.com/filipspl/bayesian-rf-knime-scikit
- Owner: filipsPL
- Created: 2019-11-08T11:12:56.000Z (about 5 years ago)
- Default Branch: master
- Last Pushed: 2019-11-08T13:09:56.000Z (about 5 years ago)
- Last Synced: 2024-11-12T13:24:59.695Z (3 months ago)
- Topics: bayesian-optimization, knime, knime-analytics-platform, python, random-forest, scikit-learn
- Language: Python
- Size: 1.09 MB
- Stars: 1
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# bayesian-rf-knime-scikit
Bayesian Optimization of RandomForest parameters, with scikit-learn, to be used in KNIME in Python learner node. Based on https://github.com/fmfn/BayesianOptimization/ by fmfn.
Prerequisities:
- `pip install bayesian-optimization`
Why?
1. Parameter Optimization Loop Node(s) doesn't work as expected for some data. Including Bayesian optimization.
2. You may want to use scikit-learn instead of KNIME or Weka implementation.
3. You can tune this workflow to optimize other parameters for many different scikit algorithms.
## Setup
- In python node please select python2.
- copy&paste the python code into the code window of Python Learner (`python-learner.py`) and Python Predictor (`python-predictor.py`)
- sample workflow:
- in the input table, the class should be in the last column
- fine tuning - edit variables at the top of the `python-learner.py`:
```python
#
# Bounded region of parameter space
#
parameterDict = { 'n_estimators': (100, 1200),
'max_depth': (5, 30),
'min_samples_split': (2, 100),
'min_samples_leaf': (1, 10)
}
#
# bayesian configuration
#
init_points = 5
n_iter = 20
```
- please note: scripts (after slight modifications) can be run from the command line
- sample data file provided (`nr-ahr-lite.csv ` from my [tox21 dataset](https://github.com/filipsPL/tox21_dataset))
## Standard output
Among some training progress data (static) info about best parameters found is displayed:
```
Best params: {'min_samples_split': 2, 'n_estimators': 205, 'max_depth': 30, 'min_samples_leaf': 1}
Best target value: 0.837006427916
```
## ROC output (ROC curve node)
