Spend Quality and Usage Analysis Tool (SQUAT)

This Project is a tool to analyse Text and categorize the text into the categories we defined during a supervised learning process to create the ML Model. We started off with Bank Statement transactions to give a comprehensive report on the spend, earning and usage of an user. It does the following job:

• Creates and trains a Machine learning model to classify transactions based on the narration.
• All the training and other repeatative work is already done for you.
• Once the package is installed with pip, the developer just need to pass the bankstatement dataframe to get the report.

Project Components

SQUAT contains the packages or libraries required for supporting and running the whole process.

1. spacy
2. Core ENG package for spacy
3. pandas
4. jupyter notebook
5. arghandler

Source:
https://github.com/binayr/SQUAT

About the ML model

Version 0.1

The model is created based on most common keyword observed from the bankstatements of singapore. This project has a large scope of improving the accuracy and adding more classifications in future depending on the type of dataset available to us.

Everytime we update the model a new version of SQUAT is supposed to get released.

Version 1.0.3

We have moved beyond the above limitation. Users can Now train the ML Model on any data set on any location they want and use the location to import the model and categorize text. There are certain rules to follow to train the ML model. Please check the below csv file and maintain the training data accordingly.

Create and use whl file

• with and updated setup.py execute the following command to create a whl file, python setup.py bdist_wheel

• Please make sure you have pre-installed pandas, spacy and jupyter from standard chartered artifactory in your virtualenv

• Also make sure once spacy is installed the eng core library is also pre-installed in the virtualenv using pip.

• Now you can pip install squat using the whl file or from standard chartered artifactory if it is hosted.

API

• You can import the utility by typing the following, from squat.Classifier.ClassifierUtil import ClassifierUtil

• Read any csv or excel using pandas and create a dataframe. Please make sure the df has the following header atleast, date, description, debit, credit, runningbalance (irrespective of the order)

• The ClassifierUtil can be initialized using the above df.

• Once initialized please make sure to call obj.evaluate() to evaluate each transaction.

• Once evaluated you can call get_analysis method to get the comprehensive analysis or call show_stat to get the statistics.

OR

• You can import the utility by typing the following, from squat.Classifier.ClassifierUtil import ClassifierUtilRaw

• Read any csv or excel using pandas and create a dataframe. Please make sure the df has the following header atleast, date, description, debit, credit, runningbalance (irrespective of the order)

• The ClassifierUtilRaw can be initialized to get the category.

• Once initialized please make sure to call obj.get_cat(text) to evaluate the category of the text.

• For Example,

   obj.get_cat('paytm transaction gurgaon')
   Out: ('Digital', 0.9632782936096191)
  

We also have few CLI Tools for training the model to our need. Developers can simply type,

squat classifier train -o <output path for the ML model> -f <path to the training data set in csv>

The above code will create the model in the path you have provided and Once done You can use the same path with ClassifierUtilRaw to start the categorization.