Datasets:
item_id large_string | instance_id large_string | channel large_string | window int64 | num_variates int64 | context list | target list | freq large_string | prediction_length int64 | term large_string | seasonality int64 |
|---|---|---|---|---|---|---|---|---|---|---|
ev0_WELL-00001_20170201010207_P-TPT_w0 | ev0_WELL-00001_20170201010207 | P-TPT | 0 | 1 | [
100.92054748535156,
100.91835021972656,
100.91615295410156,
100.91390228271484,
100.91165161132812,
100.9094467163086,
100.9072494506836,
100.9050521850586,
100.90284729003906,
100.90065002441406,
100.89845275878906,
100.89624786376953,
100.89405059814453,
100.8917465209961,
100.88954925... | [
100.22795104980469,
100.2269515991211,
100.2259521484375,
100.22489929199219,
100.22384643554688,
100.22284698486328,
100.22184753417969,
100.2208480834961,
100.2198486328125,
100.2188491821289,
100.21784973144531,
100.21685028076172,
100.21585083007812,
100.21475219726562,
100.213752746... | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201010207_P-TPT_w1 | ev0_WELL-00001_20170201010207 | P-TPT | 1 | 1 | [
100.54601287841797,
100.54601287841797,
100.5460205078125,
100.54602813720703,
100.54602813720703,
100.5460433959961,
100.5460433959961,
100.54605102539062,
100.54605102539062,
100.54605865478516,
100.54606628417969,
100.54606628417969,
100.54608154296875,
100.54608154296875,
100.5460891... | [
100.5465087890625,
100.5465087890625,
100.5465087890625,
100.54651641845703,
100.54651641845703,
100.54651641845703,
100.54651641845703,
100.54651641845703,
100.5465316772461,
100.5465316772461,
100.5465316772461,
100.5465316772461,
100.54653930664062,
100.54653930664062,
100.54653930664... | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201060114_P-TPT_w0 | ev0_WELL-00001_20170201060114 | P-TPT | 0 | 1 | [100.14689636230469,100.14689636230469,100.14689636230469,100.14689636230469,100.14689636230469,100.(...TRUNCATED) | [100.07319641113281,100.07415008544922,100.0750503540039,100.07599639892578,100.07689666748047,100.0(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201060114_P-TPT_w1 | ev0_WELL-00001_20170201060114 | P-TPT | 1 | 1 | [100.54679107666016,100.54679107666016,100.54679870605469,100.54679870605469,100.54680633544922,100.(...TRUNCATED) | [100.2800521850586,100.27784729003906,100.27555084228516,100.27335357666016,100.27114868164062,100.2(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201110124_P-TPT_w0 | ev0_WELL-00001_20170201110124 | P-TPT | 0 | 1 | [100.05184936523438,100.05394744873047,100.0560531616211,100.05815124511719,100.06024932861328,100.0(...TRUNCATED) | [99.94737243652344,99.94737243652344,99.94737243652344,99.94737243652344,99.94737243652344,99.947372(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201110124_P-TPT_w1 | ev0_WELL-00001_20170201110124 | P-TPT | 1 | 1 | [99.94924926757812,99.94925689697266,99.94925689697266,99.94925689697266,99.94927215576172,99.949272(...TRUNCATED) | [100.02279663085938,100.02169799804688,100.0206527709961,100.01954650878906,100.01844787597656,100.0(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201160311_P-TPT_w0 | ev0_WELL-00001_20170201160311 | P-TPT | 0 | 1 | [99.89456939697266,99.89363098144531,99.8926773071289,99.8917465209961,99.89080810546875,99.88986968(...TRUNCATED) | [99.74787139892578,99.74787139892578,99.74787139892578,99.74787139892578,99.74787139892578,99.747871(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170201160311_P-TPT_w1 | ev0_WELL-00001_20170201160311 | P-TPT | 1 | 1 | [99.75328063964844,99.75331115722656,99.75332641601562,99.75335693359375,99.75337982177734,99.753410(...TRUNCATED) | [99.10523223876953,99.11177825927734,99.11831665039062,99.12486267089844,99.13140106201172,99.137939(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170202020343_P-TPT_w0 | ev0_WELL-00001_20170202020343 | P-TPT | 0 | 1 | [99.14938354492188,99.14938354492188,99.14938354492188,99.14938354492188,99.14938354492188,99.149383(...TRUNCATED) | [99.0401611328125,99.04122161865234,99.04226684570312,99.04332733154297,99.04438018798828,99.0454330(...TRUNCATED) | 2s | 1,024 | long | 1 |
ev0_WELL-00001_20170202020343_P-TPT_w1 | ev0_WELL-00001_20170202020343 | P-TPT | 1 | 1 | [98.89753723144531,98.89533233642578,98.89312744140625,98.89092254638672,98.88871765136719,98.886520(...TRUNCATED) | [98.9457778930664,98.9457778930664,98.94577026367188,98.94576263427734,98.94574737548828,98.94573974(...TRUNCATED) | 2s | 1,024 | long | 1 |
IND-eval
Dataset Summary
IND-eval is a curated benchmark for evaluating time series forecasting models on real-world industrial data. It's assembled from multiple public sources, each brought in through a documented processing script.
Dataset Structure
main always reflects the latest release. Each release is also tagged
with its version, for a reproducible read pinned to that release. Each
source has its own schema, so it's loaded as its own config - see
Source Data for the list:
from datasets import load_dataset, get_dataset_config_names
ds = load_dataset("cognitedata/IND-eval", "3w") # latest release, one source
ds = load_dataset("cognitedata/IND-eval", "3w", revision="v0.1.3") # pinned release
sources = get_dataset_config_names("cognitedata/IND-eval") # every source in the latest release
all_ds = {name: load_dataset("cognitedata/IND-eval", name) for name in sources}
IND-eval/
├── data/
│ └── <source>/ # this source's processed output (one or more .parquet files)
└── raw/
└── <source>/ # only when the source's license allows redistribution
├── README.md # provenance + license notes
├── LICENSE # this source's own license
├── scripts/ # self-contained scripts that produced data/<source>/
└── data/ # the unmodified source files
Source Data
| Source | Recorded system |
|---|---|
| NEATS | Offshore drilling streams with up to 12 canonical drilling variables at 1, 5, 10, or 12-second cadence |
| 3W | Multivariate oil-well episodes organized into ten operating-event groups |
| PRONTO | Six annotated multiphase-flow transients with 12 targets and three known-future control variables |
| Mesa del Sol | Fifteen months of microgrid power, voltage, frequency, and temperature measurements |
| Tablet compression | Commercial batches with five direct-compression process targets |
| MetroPT-3 | Five channels from a metro train air-production unit |
| ETT (scripts only, no config - see below) | Two seven-channel electricity-transformer series |
| Electricity | 370 client electricity meters, treated as independent univariate series |
| Solar | Simulated power from 137 photovoltaic plants, treated as independent univariate series |
ETT's upstream license (CC BY-ND 4.0) forbids redistributing derived data, so it
ships no data/ett/ and has no config_name - get_dataset_config_names() returns
the other eight sources. Its raw/ett/scripts/ still let you reproduce it locally;
see raw/ett for how.
Considerations for Using the Data
Licensing
IND-eval itself has its own overall license, covering the compiled
dataset and the processing scripts - see LICENSE.
Each source's original raw data separately retains its own license -
see raw/<source>/LICENSE and raw/<source>/README.md.
Reproducibility
Every source's data/<source>/ is produced by re-running
raw/<source>/scripts/ against raw/<source>/data/, when the source's
license allows redistribution.
load_dataset() won't fetch raw/<source>/ for you. Download it
separately, e.g. with the hf CLI:
hf download cognitedata/IND-eval --repo-type dataset \
--include "raw/<source>/*" --local-dir .
Then, with uv installed, rerun the script:
uv run raw/<source>/scripts/prepare.py \
--raw-dir raw/<source>/data --output-dir data/<source>
Each script declares its dependencies inline (PEP 723),
pinned by the prepare.py.lock next to it - no environment setup needed.
Contributing
New sources are welcome, following the raw/<source> layout described
above. Open a pull request
on this repo - via the "Contribute" button on the Hub, or the
hf CLI's
--create-pr option if you don't have write access.
- Downloads last month
- 277