zhipenghoustat / record-screening-experiment

Record Screening Experiment

This repo contains a Python script for simulation experiment for the Paper "Enhancing Recall in Automated Record Screening: A Resampling Algorithm".

The script is a versatile tool for running experiments and checking how well the resampling algorithm from the paper works under different situations. It's set up to explore a wide range of scenarios and parameters, providing useful insights into how the algorithm behaves.

Function Definitions

• calculate_sample_size(R, c)

  Calculates required sample sizes based on specified parameters.

• sequence_gen_random(N, prevalence, seed)

  Generates a random sequence of data with a given prevalence.

• sequence_gen_simulated(N, prevalence, seed)

  Generates a simulated sequence of data with priority scores.

• sequence_gen_real()

  Retrieves real data from CSV files and ranked IDs from a serialized file.

• record_sampling(y_full, k, seed)

  Performs record sampling to select a subset of records.

• result_analysis(y_full, id_ranked, sample_list, c)

  Analyzes the sampled data to calculate recall, workload, and related metrics.

• experiment(N, prevalence, R, c, seed, round, mode)

  Conducts experiments or simulations with specified parameters and collects relevant statistics.

Modes

The script conducts experiments in different modes:

- "sim" mode involves simulating data with simulated sequences.
- "random" mode simulates random data sequences.
- "real" mode uses real data retrieved from CSV files. It loads real data from CSV files and evaluates the system's performance using the same metrics.