next-purchase-day — TsFile

This dataset is a conversion of the HuggingFace dataset alexgrigoras/next-purchase-day to the Apache TsFile format. The original CSV files are also kept here, as in the source repository.

Original dataset

• Source dataset: alexgrigoras/next-purchase-day
• Content: a GluonTS-format purchase time-series — 34 users (item_id) per split, each with a sequence of purchase times (target, seconds-of-day). Three splits: train / validation / test.

What is in this repository

data/
├── train.tsfile          # converted (TsFile)
├── validation.tsfile
└── test.tsfile
df_purchases_train.csv     # original CSVs, copied verbatim
df_purchases_validation.csv
df_purchases_test.csv

TsFile storage mapping (table model, per split)

Role	Column	Type	Notes
TAG	item_id	STRING	34 users → 34 devices
Time	within-series position index	INT64	0, 1, 2, … per series
FIELD	target	INT32	purchase second-of-day (integer, 0..86340)
FIELD	feat_static_cat_0	INT64	per-series static category

Conversion notes

• GluonTS nested rows expanded to a long table: each source row is one sequence (start, target[], item_id, feat_static_cat[1], feat_dynamic_real); the target array is exploded so each point becomes one row.
• Three splits → three separate TsFiles (train / validation / test); the original split is preserved, not merged.
• TAG = item_id (34 devices). Time = within-series position index (0, 1, 2, …). The source start field is a 1970-01-01 placeholder (not a real timestamp), so the ordinal index is used as the time axis.
• feat_static_cat → feat_static_cat_0 (INT64, the single static category).
• Dropped (with consent): feat_dynamic_real — entirely null across all splits (an empty column carrying no information). No rows dropped (train 2,448 / validation 2,584 / test 2,720 points, matching the source exactly).
• The original CSVs (df_purchases_{train,validation,test}.csv) are kept verbatim.

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/train.tsfile")
schemas = reader.get_all_table_schemas()
tname = next(iter(schemas))

cols = ["item_id", "target", "feat_static_cat_0"]
with reader.query_table(tname, cols, batch_size=65536) as rs:
    while (batch := rs.read_arrow_batch()) is not None:
        df = batch.to_pandas()
        # ... process ...
reader.close()

Citation

@misc{next_purchase_day,
  title  = {next-purchase-day},
  author = {alexgrigoras},
  url    = {https://huggingface.co/datasets/alexgrigoras/next-purchase-day},
  publisher = {Hugging Face}
}

The source HuggingFace dataset does not declare an explicit license.