Vita-Mojo/kitchensim-v1

KitchenSim v1是一个合成餐厅运营数据集，通过参数模拟器生成，包含60个模拟餐厅厨房在7天内的事件流。该数据集专为需求预测基准测试、运营状态推断和时间序列事件数据的自监督表示学习而设计。它是发布就绪版本，移除了隐藏的真实状态列，仅包含下游消费者在实际运营数据中可见的内容。数据集包含1,163,151个事件，约16 MB压缩parquet文件，涉及193,837个订单，具有可重复性。商店类型包括五种原型厨房，具有不同的运营配置文件，动态特性包括非均匀泊松到达、状态机和非线性负载因子等。使用案例包括自监督预训练、需求预测模型基准测试和运营状态推断研究。局限性包括数据是合成的、仅包含7天数据、无价格/菜单动态以及无客户侧状态。KitchenSim v1 is a synthetic restaurant operations dataset, generated by a parametric simulator, containing an event stream for 60 simulated restaurant kitchens across 7 days. It is designed for benchmarking demand forecasting, operational-state inference, and self-supervised representation learning on time-series event data. This is the publish-ready version, with the hidden ground-truth state column stripped, including only what a downstream consumer would see in real operational data. The dataset comprises 1,163,151 events, approximately 16 MB compressed parquet file, with 193,837 orders generated, and offers reproducibility. Store types include five archetypal kitchens with distinct operational profiles, and dynamics feature non-homogeneous Poisson arrivals, state machines, and non-linear load factors. Use cases include self-supervised pretraining, demand-forecasting benchmarking, and operational state-inference research. Limitations include being synthetic, having only 7 days of data, lacking price/menu dynamics, and no customer-side state.