mysteriousauthor/spikeprophecy-steinmetz

这是一个处理后的数据集，名为SpikeProphecy Steinmetz (processed)，源自2019年公开的Steinmetz数据集。它包含50毫秒分箱的尖峰计数张量，作为SpikeProphecy基准测试（NeurIPS 2026数据集和基准赛道）的主要数据基础。数据集通过确定性预处理流程生成，包括读取原始尖峰时间、以50毫秒为间隔分箱成整数计数向量、过滤低平均放电率（低于0.1赫兹）的神经元单元，并输出每个会话的[n_units, n_bins]张量。每个会话使用70/15/15的训练/验证/测试时间分割，分割边界存储在元数据中。数据集包含39个会话文件（每个文件对应一个会话）和一个元数据文件，用于记录会话信息和分割细节。数据格式为uint8，最大观察值为43，无实际裁剪。该数据集用于时间序列预测任务，特别是神经群体尖峰预测，支持神经科学研究。Processed 50 ms-binned spike-count tensors derived from the publicly released Steinmetz 2019 dataset, used as the primary substrate of the SpikeProphecy benchmark (NeurIPS 2026 Datasets & Benchmarks track). This dataset is a deterministic preprocessing of the Steinmetz 2019 recordings, involving binning spike times at 50 ms intervals into integer count vectors, dropping units with mean firing rates below 0.1 Hz, and producing per-session tensors with a 70/15/15 train/val/test temporal split. It includes 39 session files and metadata, designed for autoregressive neural population forecasting tasks.