uID3 w Pythonie

Uncertain Decision Tree Classifier

Celem prac jest implementacja mechanizmu budowania drzew decyzyjnych z danych niepewnych.

• Prace bazowac beda na artykule: Uncertain Decision Tree Classifier for Mobile Context-Aware Computing
• Prace bazowac beda na kodzie: UID3
• Prace powinny implementowac klasyfikator zgodnie z konwencja sklearn
• Prace maja finalnie potwierdzic rownowaznosc implementacji Java i Python (z wylaczeniem sytuacji gdzie wersja Python poprawialaby bledne dzialanie wersji Java)
• Do implementacji polecam wykorzystac framework nbdev
• Implementacja powinna umozliwiac rysowanie drzew z wykorzystaniem graphviz:
• Implementacja powinna umozliwiac eksport drzew do formatu HMR+, ktory zaczytywany jest przez webowy edytor HWED

• Kod powinien byc napisany i udokumentowany zgodnie z PEP8

Install

pip install uid3

How to use

from pyuid3.data import Data
from pyuid3.uid3 import UId3
from pyuid3.uncertain_entropy_evaluator import UncertainEntropyEvaluator

data = Data.parse_uarff("../resources/weather.nominal.uncertain.arff")
uid3 = UId3()
tree = uid3.fit(data, entropyEvaluator=UncertainEntropyEvaluator(), depth=0)

instance = data.instances[0]

prediction = uid3.predict(instance)
print(prediction)

no[0.6]

How to contribute

Please, read the official tutorial first.

1. Clone the repository and install requirements.
2. Start jupyter notebook in project's folder and open it in your browser.
3. Click on New -> Terminal in the right upper corner.
4. First, run nbdev_install_git_hooks to avoid conflicts.
5. Apply changes to notebooks in ./src folder.
6. Build library using nbdev_build_lib command. This will create pyuid3 folder in project's root directory.
7. To be able to build documentation you must make sure that previously created pyuid3 folder is also present (and updated!) in ./src directory. Therefore, you should go to .src/ folder and create simlink to pyuid3 by executing the following command: 'ln -s ../pyuid3 pyuid3'. Please, make sure that simlink was created properly. In my case, I have to delete the ./src/pyuid3 folder first (if it already exists) before creating simlink to pyuid3, otherwise it will not be properly updated.
8. Now you should be able to build documentation using nbdev_build_docs command. If any errors occured, you have to fix them first. The error we were often facing was caused by nested imports i.e. when class A imports class B, and class B imports class A.
9. Commit your changes and push to gitlab to see if the page can be deployed.

Issues

• Nested imports (i.e. when class A imports class B, and class B imports class A) are causing errors.
• Some notebooks' documentation is looking different than the others', despite all source notebooks having the same structure. See: Data, DataScrambler, Instance, ParseException, UId3, UncertainEntropyEvaluator.
• In order for UId3 class to be consistent with sklearn estimators, in fit method, Data object 'data' should be split into array-like 'X' and 'y' and then fit should be rewritten as 'fit(X, y)'. Similarly, predict method should take array-like 'X', not Instance object. However, I think these improvements would require many changes not only in algorithm implementation but in the entire project's structure in general.