STESH: Attention Weights Across 200 Training Epochs

151,510 attention weights tracked over 200 epochs for the STESH spatial transcriptomics clustering model. The dataset, authored by Huihui Zhang and last updated in June 2026, is shared under a CC-BY-4.0 license on figshare. It supports the evaluation of a method that integrates gene expression, spatial location, and histology for tissue analysis.

Use Cases

• Analyzing the convergence behavior of attention mechanisms based on the 200-epoch training run
• Benchmarking multi-view graph neural network architectures based on the described expression, histology, and spatial modules
• Studying model interpretability in bioinformatics based on the provided attention weights
• Reproducing or validating the STESH clustering method's performance metrics mentioned in the description

Strengths

• Contains 151,510 specific data points for analysis.
• Covers a complete training cycle of 200 epochs.
• Shared under a permissive CC-BY-4.0 license for reuse.

Limitations

• Row count and column-level documentation are absent; field semantics must be inferred after download.
• The 23.5 KB file size indicates a very limited scope, likely containing only model metadata or summary statistics.

Provenance

Source

Huihui Zhang via figshare

Collection Method

Likely generated during the training and evaluation of the STESH spatial transcriptomics clustering model.

Freshness

Last updated 2026-06-01 17:40:33; freshness should be verified.