fairlyuncertain 0.0.4

Heteroscedastic uncertainty estimates for fair algorithms.

FairlyUncertain Documentation

Table of Contents

• How to Add a New Dataset
• How to Write a New Experiment
• How to Add a New Model/Algorithm
• How to Add a New Metric/Evaluation

How to Add a New Dataset

To add a new dataset, follow these steps.

1. Import Necessary Libraries

   Ensure you have the required libraries imported:

   import pickle
   import os
   import numpy as np
   import pandas as pd
   import requests
   

2. Create Dataset Loader Function

Implement a function to load your new dataset. Here is an example:

def get_new_dataset():
    # Add logic to download, preprocess and return your dataset
    pass

3. Update Data Loaders Dictionary

Add the new dataset loader function to the dictionary of data loaders:

dataloaders = {
    'New Dataset': get_new_dataset,
    # other datasets...
}

4. Ensure Dataset Caching

Make sure your dataset is cached for quick access:

def cache_dataset(name):
    # Logic to cache the dataset
    pass

def read_dataset(name):
    # Logic to read the cached dataset
    pass
  ```
5. **Load Dataset Instance**

Ensure that the dataset can be loaded correctly:

```python
def load_instance(name, train_split=.8):
    # Logic to load dataset instance
    pass

How to Write a New Experiment

To write a new experiment, follow these steps:

1. Import Necessary Libraries

Ensure all required libraries are imported:

from tqdm import tqdm
import numpy as np
...
import fairlyuncertain as fu

2. Define Algorithms to Test

List the algorithms you want to test:

algorithm_names = ['Algorithm1', 'Algorithm2']
algorithms = {name: fu.algorithms[name] for name in algorithm_names}

3. Run Experiments

Loop through datasets and algorithms to run your experiments:

results = {}
for dataset in tqdm(yp.datasets):
    instance = fu.load_instance(dataset)
    results[dataset] = {'instance': instance}
    for algo_name in algorithms:
        results[dataset][algo_name] = algorithms[algo_name](instance)

4. Evaluate and Plot Results

Evaluate the results using the appropriate metrics and plot the outcomes:

for metric_name in fu.metrics:
    fu.plot_results(results, algorithms, fu.datasets, metric_name=metric_name)

How to Add a New Model/Algorithm

To add a new model or algorithm, follow these steps:

1. Add your Model Class if Necessary

If your new algorithm requires a new model class, make those changes:

class Model:
    def __init__(self, **kwargs):
        # Initialization logic
        pass

    def fit(self, X, y):
        # Fit logic
        pass

    def predict(self, X):
        # Prediction logic
        pass
  ```

2. **Import Necessary Libraries**

Ensure necessary libraries and modules are imported:

```python
import numpy as np
...
import fairlyuncertain as fu

3. Define the New Algorithm

Implement the logic for your new algorithm:

def new_algorithm(instance, Model):
    # Algorithm implementation
    pass

4. Add Algorithm to Dictionary

Update the algorithms dictionary to include your new algorithm:

algorithms = {
    'New Algorithm': new_algorithm,
    # other algorithms...
}

How to Add a New Metric/Evaluation

To add a new metric or evaluation, follow these steps:

1. Import Necessary Libraries

Import required libraries for your metric:

import numpy as np
...

2. Define the New Metric

Implement the logic for the new metric:

def new_metric(pred, y, group):
    # Metric calculation logic
    pass

3. Add Metric to Dictionary

Update the metrics dictionary to include your new metric:

metrics = {
    'New Metric': new_metric,
    # other metrics...
}

4. Use New Metric in Evaluation

Ensure the new metric is used in the evaluation process:

def evaluate_metrics(results, metrics):
    for metric_name in metrics:
        for dataset in results:
            results[dataset][metric_name] = metrics[metric_name](
                results[dataset]['pred'],
                results[dataset]['y'],
                results[dataset]['group']
            )