API Reference

This section provides detailed documentation for all classes and functions in scDataset.

Main Dataset Class

scDataset

Iterable PyTorch Dataset for on-disk data collections with flexible sampling strategies.

Multi-Modal Data Support

MultiIndexable

Container for multiple indexable objects that should be indexed together.

Transform Functions

adata_to_mindex	Transform AnnData/AnnCollection batch to MultiIndexable with optional obs columns.
hf_tahoe_to_tensor	Transform HuggingFace Tahoe-100M sparse gene expression data to dense tensors.
bionemo_to_tensor	Fetch callback for BioNeMo SingleCellMemMapDataset.

Sampling Strategies

SamplingStrategy	Abstract base class for sampling strategies.
Streaming	Sequential streaming sampling strategy with optional buffer-level shuffling.
BlockShuffling	Block-based shuffling sampling strategy.
BlockWeightedSampling	Weighted sampling with block-based shuffling.
ClassBalancedSampling	Class-balanced sampling with automatic weight computation.

Experimental Features

Warning

Features in the experimental module are subject to change and may be modified significantly or removed entirely in future releases.

suggest_parameters	Suggest optimal parameters for scDataset based on system resources.
estimate_sample_size	Estimate the memory size of a single sample from the data collection.