NaiveBayesClassifier_Python

Naive Bayes Classifier agent built in Python 3

About

Naive Bayes classifier is a probabilistic classifier that uses Bayes' theorem to estimate the classification of a data entity (row of data). It depends on a (naive) assumption of independence between the probability of effects given a cause to estimate the probability of a cause given multiple effects, hence the name Naive Bayes classifier.

Running

• NaiveBayesClassifier.py ./NaiveBayesCalculator.py This is the "main" function that runs the program. To run the program enter:

  $ python NaiveBayesClassifier.py "./absolute/path/to/data.csv" <"./absolute/path/to/estimation_data.csv">

  For example, running the command:

  $ python NaiveBayesClassifier.py "./data/pima-indians-diabetes.csv"

  will run the Naive Bayes agent on the pima-indians-diabetes.csv dataset. While this may not do anything interesting, going into the main log, (./logs/main.log) the last line will display the accuracy of the classifier on the passed dataset. A more interesting example, however:

  $ python NaiveBayesClassifier.py "./data/primate-factors.csv" "./data/primate-factors-no-class-var.csv"

  will create a new .csv file, ./data/primate-factors-predictions.csv and a new .txt file, ./data/primate-factors-predictions-about.txt that describes the unknown dataset with estimated classes and the accuracy of the classifier on the known dataset, respectively.

  The program assumes data is in the following format:

  | AttributeR,C | AttributeR,C | ... | AttributeR,C | Class VariableR |

  | :------------: | :------------: | :---: | :------------: | :---------------: |

  | Attribute1,1 | Attribute1,2 | ... | Attribute1,N | Class Variable1 |

  | Attribute2,1 | Attribute2,2 | ... | Attribute2,N | Class Variable2 |

  | ... | ... | ... | ... | ... |

  | AttributeN,1 | AttributeN,2 | ... | AttributeN,N | Class VariableN |

Versions & Revisions

• Version 1.0, last modified 10/27/2016 (current)
  • Base implementation
  • Probability is only estimated on a Normal (Gaussian) Distribution
  • Known data is currently split at 67% training, 33% testing