Skip to content

README.md

Latest commit

 

History

History
96 lines (81 loc) · 2.19 KB

README.md

File metadata and controls

96 lines (81 loc) · 2.19 KB

DataPro algorithm

The task is learning the common structure of data. The DataPro algorithm is the algorithm that finds statistically significant patterns in a set of token sequences. This repository is Implemented from the paper "Learning the Common Structure of Data" by Kristina Lerman and Steven Minton.

Installation

pip install datapro-learning

Get started

How to use model with this lib:

from datapro import DataPro
import pandas as pd

# Read data file
df = pd.read_excel("street_road.xlsx")
df = df.dropna()

# Choose a column, type list or pandas series.
data_sample = df["หน่วยรับผิดชอบ"]


# Create datapro object.
datapro = DataPro(alpha=0.05, k_percentage=10)

# Train with data
datapro.fit(data_sample)

# show result
print(datapro.evaluate_score())

Customize

Load new type's tree.

This algorithm uses a type's tree to assign types to tokens and can be configured by using JSON file with a structure like the below.

NOTE: When there is a new significant token, it's will be a child of these nodes as a specific node.

  {
    "TOKEN": {
      "regex": ".*",
      "children": [
        "PUNCT",
        "ALPHANUM"
      ],
      "parent": ""
    },
    "PUNCT": {
      "regex": "^[\\.\\?!,:'()\"]$",
      "children": [],
      "parent": "TOKEN"
    },
    "ALPHANUM": {
      "regex": "[\\da-zA-Z]+",
      "children": [
        "ALPHA",
        "NUMBER"
      ],
      "parent": "TOKEN"
    },
    "ALPHA": {
      "regex": "^[a-zA-Z]+$",
      "children": [
        "CAPS",
        "LOWER",
        "ALLCAPS"
      ],
      "parent": "ALPHANUM"
    },
  }

By default, these are all general token on type's tree.

                  TOKEN
                /        \
           PUNCT        ALPHANUM  
                     /           \
                ALPHA            NUMBER  
             /    |    \         |        \
       ALLCAPS   CAPS   LOWER   DECIMAL   INT
                                           |
                                         DIGIT

Load new file

Using method load_tree_type

  datapro.load_tree_type("<name>.json")

Reference

https://www.aaai.org/Papers/AAAI/2000/AAAI00-093.pdf