araxia 0.0.2

A minimal neural network library built from scratch for time series forecasting and analysis.

🕸️ Araxia

Araxia is a lightweight, customizable time series forecasting model tailored for structured datasets like grocery store sales. Inspired by the structure of spider webs and powered by neural networks, Araxia makes forecasting intuitive, modular, and fun.

“Pattern is the web. Prediction is the spider.” — Araxia Philosophy

---

🧠 Features

• 🔁 Lagged features with flexible windowing
• 📅 Time-aware feature engineering (cyclic time, dates, day-of-week)
• 🕸️ Neural network forecaster (built with NumPy from scratch)
• 🧪 Easy backtesting on training data
• 📈 Built-in plot support for forecasts
• 🧩 Modular: plug in your own preprocessing or model layers
• ✅ Lightweight with no heavy dependencies

---

🚀 Installation

pip install araxia

Or if you are developing locally:

git clone https://github.com/your-username/araxia.git
cd araxia
pip install -e .

if You want to run the locally installed dev environment

pip install -e .

📦 Quickstart

from araxia import create_lagged_dataset, create_dataset_with_cyclic_features
from araxia import AraxiaForecaster

Load your time series data (must include a datetime column and target)

df = your_dataframe[['date', 'sales']]

Create features

df_feat = create_lagged_dataset(df, target='sales', lags=[1, 7, 14])
df_feat = create_dataset_with_cyclic_features(df_feat, date_col='date')

Train-test split

train = df_feat.iloc[:-30]
test = df_feat.iloc[-30:]

Fit model

model = AraxiaForecaster(hidden_sizes=[16, 8], epochs=500, learning_rate=0.01)
model.fit(train.drop(columns='sales'), train['sales'])

Predict

forecast = model.predict(test.drop(columns='sales'))

Plot

model.plot(train['sales'], forecast, title="Araxia Forecast")

🧰 Core Modules

Module	Purpose
create_lagged_dataset	Generate lag features (e.g., 1-day, 7-day lags)
create_dataset_with_cyclic_features	Add sine/cosine encodings for day/month/year
create_dataset_with_date_features	Add year, month, day columns
create_dataset_with_onehot_dow	One-hot encode day-of-week
AraxiaForecaster	Custom NumPy-based feedforward network

⚙️ Customize

Araxia is modular! Replace or enhance the following:

• Use your own preprocessing logic
• Swap out the NumPy model for PyTorch/TensorFlow
• Add early stopping, dropout, or regularization
• Integrate into a pipeline with sklearn

🧪 Tests

pytest

📄 License

MIT License

🤝 Contributing

Contributions welcome! Fork the repo, create a branch, and submit a PR.