📦 dwrappr

A lightweight and extensible Python package for managing data, tailored for researchers working with structured data. In addition to general data management features, the package introduces a data structure specifically optimized for ML research. This common format enables researchers to efficiently test new algorithms and methods, streamlining collaboration and ensuring consistency in data management across projects.

🧩 Features

• 🗃️ Consistent dataset object structure for handling structured data in ML use cases
• 🔄 Support for building a file-based internal dataset collaboration platform for researchers
• 🧰 General utilities for managing data like saving and loading

🚀 Quickstart

For executing the quickstart examples and get an overview of dwrappr's functionalities, please have a look at IEEE_examples.

Additional functionalities are showcased in:

• loading_dataset_from_file.py: Shows how to load a dataset from an existing dataset file
• scanning_folder_for_datasets.py: Shows how to scann a folder vor available datasets
• dataset_functionalities.py : Shows some of the main functionalities of the DataSet class.

👀 Functionality Ipnsights

Scan folder for dataset

DATASET_FOLDER = "./data/datasets/"
available_datasets = DataSet.get_available_datasets_in_folder(
    DATASET_FOLDER
)
available_datasets.T

Loading specific dataset

DATASET_FILEPATH = "./data/datasets/manufacturing_process_ds.joblib"
ds = DataSet.load(DATASET_FILEPATH)

Generating dataset from raw data

RAW_DATA_FILEPATH= "./data/raw_data.csv"
#load raw data into pandas.DataFrame
df = pd.read_csv(RAW_DATA_FILEPATH)
"""
<some manual dataset preprocessing steps
like dropping missing values and chaning dtypes>
"""
#define metaData
meta = DataSetMeta(
    name = "example_dataset",
    synthetic_data=True,
    time_series=False,
    feature_names=["feature"],
    target_names=["target"]
)
#generate DataSet
ds = DataSet.from_dataframe(
    df=df,
    meta=meta
)
#saving dataset
ds.save("./data/example_dataset.joblib", drop_meta_json=True)

Split dataset

(train/test-split)

import numpy as np
n_instances = 100
# Create the 'product_id' feature with 3 different categorical values
product_ids = np.random.choice(['1001', '2002', '3003', '4004', '5005', '6006', '7007'], size=n_instances)
# Generate two additional numeric features
feature_1 = np.random.rand(n_instances) * 100  # Random numbers between 0 and 100
feature_2 = np.random.rand(n_instances) * 50   # Random numbers between 0 and 50
# Generate a numeric target
target = feature_1 * 0.5 + feature_2 * 0.3 + np.random.randn(n_instances) * 5  # Adding some noise
# Create a DataFrame
df = pd.DataFrame({
    'product_id': product_ids,
    'feature_1': feature_1,
    'feature_2': feature_2,
    'target': target
})

ds = DataSet.from_dataframe(
    df=df,
    meta = DataSetMeta(
        name = "example_dataset",
        synthetic_data=True,
        time_series=False,
        feature_names=["product_id", "feature_1", "feature_2"],
        target_names=["target"]
    )
)

train_ds, test_ds = ds.split_dataset(
    first_ds_size=0.5,
    shuffle=True,
    group_by_features=["product_id"]
)

📄 Help

See Documentation for details.

🛠️ Package Installation

• pip install dwrappr: grundlegende Datenverwaltungsfunktionen
• pip install "dwrappr[ml]": zusätzliche ML-spezifische Funktionen

(keep package updated with pip install dwrappr --upgrade)

🔧 Maintainer

This project is maintained by Nils