Skip to content

Latest commit

 

History

History
52 lines (38 loc) · 2.12 KB

README.md

File metadata and controls

52 lines (38 loc) · 2.12 KB
Raw

irt_test

Package for calc difficult of test tasks and ability of test subjects by Rasch model (IRT 0).

Installation

You can install irt_test using pip:

pip install irt-test

Example Usage

ℹ️ Important

🔥 Note: The Rasch model works only with binary data, also no column or row should contain a single value, for example, a row of zeros or ones will be excluded from the analysis.

Input DataFrame example:

(The 5th task will be automatically deleted)

Task 1 Task 2 Task 3 Task 4 Task 5
Julia 1 0 1 1 1
Ivan 1 1 1 0 1
Anna 1 0 0 0 1
Peter 0 0 0 1 1

For get logits of tasks and subjects use 'irt' function.

This function calculates scores for the subjects' abilities and tasks' difficulty in the form of logits using the IRT model.

Parameters:

  • df (DataFrame): A matrix with only zeros and ones values.
  • steps (int): Number of learning steps (if 0, the model will run until the error > accept).
  • accept (float): Acceptable error value (ignored when steps > 0).

Returns:

  • result (IrtResult): An object containing logit vectors, rejected subjects and tasks, and model error.
from irt_test.irt import irt

# Get logits and additional info from IRT0 model
irt_result = irt(df)

irt_result object contains logits of tasks and subjects from the IRT0 model, along with error and rejected units.

Attributes:

  • abilities (pandas.Series): Logits of subjects' abilities (subjects are the index of the Series).
  • difficult (pandas.Series): Logits of task difficulty (tasks are the index of the Series).
  • err (float): Metric of the difference between real test results and estimated results by logits.
  • rejected_tasks (list): Names of tasks that cannot be used to calculate difficulty logits.
  • rejected_subjects (list): Names of subjects that cannot be used to calculate ability logits.