Hyperparameters for Protein Model Training Across Five Datasets

A table listing hyperparameter configurations used for training neural network models on five protein datasets. The data includes batch sizes, learning rates, layer counts, and node counts for NN, CNN, and GCN architectures, with separate indications for mutational or positional splits. The dataset was authored by Shrishti Barethiya and last updated on March 25, 2026.

Use Cases

• Benchmarking model performance based on reported hyperparameter combinations like batch size and learning rate.
• Designing new experiments for protein prediction tasks based on the documented layer and node configurations.
• Analyzing the impact of different data splits (mutational/positional) on hyperparameter selection.
• Comparing architectural choices (NN, CNN, GCN) based on their specified parameters like filter sizes and graph thresholds.

Strengths

• Explicitly lists specific hyperparameter values: batch sizes (32, 64, 128, 1024, 2048), learning rates (0.005, 1e-3, 1e-4).
• Documents configurations for multiple model types: NN (dense layers, nodes), CNN (convolutional layers, filters), GCN (graph thresholds).
• Indicates separate hyperparameters for mutational or positional splits where applicable.
• Released under a permissive CC-BY-4.0 license.

Limitations

• Row count and column-level documentation are unknown, which may limit suitability assessment.
• The dataset is very small (26.5 KB), indicating a limited scope focused on parameter listings.
• Description metadata is limited; actual data structure and quality require manual inspection after download.

Provenance

Source

figshare

Collection Method

Likely compiled from experimental training runs.

Time Range

null

Freshness

Last updated 2026-03-25 17:44:43; freshness should be verified.

Geography

null