docassemble-openlcbr

A docassemble package for case outcome prediction using the analogical reasoning features of openlcbr.

This post is a primer on what an analogical reasoner does, and why it's important to have one in the open-source toolkit.

This post provides a full explanation of how the IBP algorith, which docassemble-openlcbr implements, works.

Requirements

• docassemble

Installation Procedure

Use the docassemble package manager to add the package from https://github.com/Gauntlet173/docassemble-openlcbr

Usage

Load the db_builder.yml interview for an interview that will assist you in building your own analogical reasoning tools for use with docassemble-openlcbr. That interview will provide you with a leave-one-out accuracy rating when the interview is complete.

In order to debug your reasoner, use the deep_reasoner_test.yml interview to view the detailed results of all leave-one-out tests of the prediction algorithm.

Once you have a working reasoner, implement it in your docassemble interview by following these steps:

1. Upload your reasoner file into the sources directory in docassemble.
2. Include .ibp_data in the modules block of your interview.
3. Include the following lines in the objects block of your interview, replacing reasoner_name.yml with the name of the file you uploaded:

objects:
  - reasoner: DAIBPData
  - database: DAStaticFile.using(filename="data/sources/reasoner_name.yml")
  - test_case: DAIBPCase

4. In your interview, populate the DAList test_case.factors with the factor IDs from your reasoner that are relevant to your test case.
5. Call reasoner.load(database) to initialize the reasoner. If you do not want to include the cases listed in your database, you can call reasoner.load_model_only(database), and then cases can be added to the reasoner using reasoner.add_precedent_case(case) where case is a DAIBPCase object with attributes of .id, .winner (either 'p' or 'd'), and .factors as in a test case.
6. Call reasons = reasoner.predict(test_case, issue="id of root issue"), replacing "id of root issue" with the id you gave to the root issue in your database. It will return a DATree object which is the explanation of the result.
7. Obtain the result of the prediction by looking at reasons.prediction. The value 'p' indicates that the outcome was predicted for the plaintiff, 'd' for the defendant, and 'a' indicates that the reasoner abstained. To display a user-friendly version call prediction_word(reasons.prediction), but be aware that what the reasoner considers 'plaintiff' may represent 'true' in your interview, so prediction_word is not always appropriate.
8. To display the reasons to the user, you must include the following in the features block of your interview:

features:
  javascript: docassemble.openlcbr:data/static/list_collapse.js
  css: docassemble.openlcbr:data/static/list_collapse.css

9. Then, use reasons.display_tree() to display a collapsing tree interface of the reasoner's results.

Demos

There are four live demos of docassemble-openlcbr functionality available. The source code for all the demos is included in the package:

0. The Trade Secrets demo is a minimal implementation of docassemble-openlcbr.
1. The AIP Tool demonstrates a real-world use of an analogical reasoner built using docassemble-openlcbr, and display's the reasoner's reasons on the last page of the interview.
2. The Reasoner Builder can be used to create a reasoner database and get a raw leave-one-out predictive score.
3. The Deep Reasoner Tester can be used to see the reasons of all the leave-one-out tests in order to troubleshoot bad predictions.

Clio Integration Demo

Thanks to Clio's sponsorship of the ABA Fellowship Project, there is also a version of the Trade Secret interview which uses information obtained from a live Clio account, demonstrating the possibility of integrating docassemble-openlcbr with live data acquired over the Clio API.

In this demo, both the information from the test case and the precedents are obtained from the live Clio account. Only the factors and the issue model need to be specified in the database.

Click here for a live demo of the integration between Clio and docassemble-openlcbr.

Documentation

A number of blog posts about the project are available at Jason Morris' Medium page.

Documentation for docassemble-openlcbr is still a work in progress, and will be available at https://gauntlet173.github.io/docassemble-openlcbr/.

Help

Please report issues at the docassemble-openlcbr GitHub.

For support, join the #analogyproject channel at the docassemble slack.

To offer feedback, contact Jason Morris.

Thanks

• Kevin Ashley and Stefanie Bruninghaus for developing the IBP Algorithm
• Matthias Grabmair for the Python implementation of IBP in openlcbr on which docassemble-openlcbr is based. (And for re-implementing it in Python 2.7!)
• The American Bar Association Center for Innovation, and in particular Chase Hertel, Sarah Glassmeyer, and Joshua Furlong, who were particularly involved in selecting my proposal.
• Jack Newton, Joshua Lenon, and Chris Thompson of Clio, who were instrumental in my fellowship receiving sponsorship funding from Clio.
• Jonathan Pyle, for developing docassemble, and for his constant and expert assistance over the three months that this project was underway.
• The docassemble user community who were extremely helpful and patient with a lawyer who was learning docassemble and Python.

Version History

• 0.5.0 - First "feature complete" version.
• 0.5.3 - Fixed - Periods of Separation Not Working
• 0.5.4 - Fixed - Problems adding and editing factors