Add datasets / benchmarks module

[RaczeQ]

The library focuses on creating embeddings for downstream tasks, but we have a hard time benchmarking different embedders and models.

We would like to have example datasets that could be used for benchmarking the models. This module could be connected to HuggingFace API and download datasets from Kraina directory.

Example datasets to be included:

• AirBnB renting costs (with property features and prices) - point dataset that could be aggregated to spatial index
• Bike sharing stations (predict whether hex contains the station or not) - heavily unbalanced dataset, could be represented as binary or distance to nearest stations hex.
• Find some Iris dataset equivalent for people who are new to the geospatial domain - easy to grasp and very easy to predict. with simple CountEmbedder.
• Maybe some population dataset in the form of Eurostat squares that can be interpolated into H3 hexes with some clever mapping over buildings?
• Some intricate metric/index, for example calculate urban complexity index from https://github.com/winstonyym/urbanity (already calculated dataset: https://www.nature.com/articles/s41597-023-02578-1) and we could see if simple model with less data can somehow reproduce those values. Other repositories: https://github.com/PengyuanLiu1993/XAI-Urban-Analytics,

[Calychas]

1. Airbnb X -> points, region ids for different resolutions (h3?), airbnb features Y -> target (mean rent cost, ...)

2. BSS X -> region ids for different resolutions Y -> is there a bike station in the region