Dataset Card for "kdd210_hourly"

More Information needed

Download the Dataset:

from datasets import load_dataset

dataset = load_dataset("LeoTungAnh/kdd210_hourly")

Dataset Card for Air Quality in KDD cup 2018

Originally, the dataset is from KDD cup 2018, which consists of 270 time series data with different starting time. This dataset encompasses 210 hourly time series data points starting from 2017-01-01T14:00:00. The dataset reveals the air quality levels in 59 stations in 2 cities from 01/01/2017 to 31/03/2018.

Preprocessing information:

• Grouped by hour (frequency: "1H").
• Applied Standardization as preprocessing technique ("Std").
• Preprocessing steps:
  1. Standardizing data.
  2. Replacing NaN values with zeros.

Dataset information:

• Missing values are converted to zeros.
• Number of time series: 210
• Number of training samples: 10802
• Number of validation samples: 10850 (number_of_training_samples + 48)
• Number of testing samples: 10898 (number_of_validation_samples + 48)

Dataset format:

  Dataset({
  
      features: ['start', 'target', 'feat_static_cat', 'feat_dynamic_real', 'item_id'],
      
      num_rows: 210
      
  })

Data format for a sample:

• 'start': datetime.datetime

• 'target': list of a time series data

• 'feat_static_cat': time series index

• 'feat_dynamic_real': None

• 'item_id': name of time series

Data example:

{'start': datetime.datetime(2017, 1, 1, 14, 0, 0),
 'feat_static_cat': [0],
 'feat_dynamic_real': None,
 'item_id': 'T1',
 'target': [ 1.46812152,  1.31685537,  1.26169969, ...,  0.47487208, 0.80586637,  0.33006964]
}

Usage:

• The dataset can be used by available Transformer, Autoformer, Informer of Huggingface.
• Other algorithms can extract data directly by making use of 'target' feature.