Hugging Face's logo

Datasets:
smilegeng
/
fev_datasets

Tasks:
Time Series Forecasting
Modalities:
Time-series
Sub-tasks:
univariate-time-series-forecasting
multivariate-time-series-forecasting
Size:
100K<n<1M
ArXiv:
Tags:
forecasting
tsfile
format:tsfile
timeseries
License:
Dataset card Files
xet
 Community
fev_datasets / rohlik_sales /README.md
smilegeng's picture
smilegeng
chore: translate subset README and add format:tsfile tag
52e9ab9 verified
preview code
|
Raw
History Blame Contribute Delete
3.82 kB
metadata 
license: other
task_categories:
  - time-series-forecasting
tags:
  - time-series
  - tsfile
  - format:tsfile
pretty_name: rohlik_sales (TsFile format)

rohlik_sales (TsFile format)

Source and Citation

This dataset was converted from external source data into a unified forecasting format and then into TsFile. Licensing and citation requirements follow the original source. We do not claim ownership of the original data. Unless otherwise stated, the data is intended for research use.

Dataset Statistics

Frequency Number of series Median length Observations Dynamic columns Static columns Files
1D 5,390 1,046 74,413,935 15 7 1D/1D_1..1D_5.tsfile (5 shards)
1W 5,243 150 10,516,770 15 7 1W/1W.tsfile

TsFile Storage Model

  • Each original series (id) is stored as one TsFile device (TAG dimension).
  • Static covariate columns are also stored as TAGs (device metadata): product_unique_id, name, L1_category_name_en, L2_category_name_en, L3_category_name_en, L4_category_name_en, warehouse.
  • Time-varying targets and dynamic covariates are stored as measurements (FIELDs).
  • timestamp is mapped to Time as INT64 milliseconds.
  • Table names: rohlik_sales_1D, rohlik_sales_1W.

Column Schema

Column Role TsFile type
Time Time column INT64
id TAG (device dimension) STRING
product_unique_id TAG (device dimension) DOUBLE
name TAG (device dimension) STRING
L1_category_name_en TAG (device dimension) STRING
L2_category_name_en TAG (device dimension) STRING
L3_category_name_en TAG (device dimension) STRING
L4_category_name_en TAG (device dimension) STRING
warehouse TAG (device dimension) STRING
total_orders FIELD (measurement) FLOAT
sales FIELD (measurement) FLOAT
sell_price_main FIELD (measurement) FLOAT
availability FIELD (measurement) FLOAT
type_0_discount FIELD (measurement) FLOAT
type_1_discount FIELD (measurement) FLOAT
type_2_discount FIELD (measurement) FLOAT
type_3_discount FIELD (measurement) FLOAT
type_4_discount FIELD (measurement) FLOAT
type_5_discount FIELD (measurement) FLOAT
type_6_discount FIELD (measurement) FLOAT
holiday FIELD (measurement) FLOAT
shops_closed FIELD (measurement) FLOAT
winter_school_holidays FIELD (measurement) FLOAT
school_holidays FIELD (measurement) FLOAT

Note: 10633 original id values contained invalid identifier characters and were normalized to valid device names, for example 0→_0, 1→_1, 2→_2.

Conversion Notes

  • Each original row stores one complete series (id, timestamp[], target and dynamic arrays). The converter flattens these nested rows into a long table before writing TsFile.
  • Numeric types are inferred from the source columns, for example float32 to FLOAT, float64 to DOUBLE, integers to INT64, and bool to BOOLEAN.
  • Time precision: milliseconds (INT64).
  • Large tables may be automatically sharded as <name>_1.tsfile, <name>_2.tsfile, and so on. These shards belong to the same logical table.

Reading Example 

from tsfile import TsFileReader

reader = TsFileReader("1D/1D.tsfile")
schemas = reader.get_all_table_schemas()
# Table name: rohlik_sales_1D; see \"Column Schema\" below.