etna.auto.Pool#

class Pool(value)[source]#

Bases: Enum

Predefined pools of pipelines.

Pools are divided into types by frequency (D (daily), H (hourly), MS (monthly), W (weekly), no_freq) and duration (super_fast, fast, medium and heavy).

Division by frequency:

  • The D group of pools should be chosen in a case of “D” frequency.

  • The H group of pools should be chosen in a case of “h” frequency.

  • The MS group of pools should be chosen in a case of “ME” and “MS” frequencies.

  • The W group of pools should be chosen in a case of following frequencies: “W”, “W-MON”, “W-SUN”, …, etc.

  • The no_freq group of pools should be chosen in a case of other frequencies, frequencies with the gap between timestamps more than 1 (T, Q, 2D, 4H, …, etc) and series without datetime timestamps.

Frequency in your data is defined by value of freq parameter passed to TSDataset constructor or TSDataset.create_from_misaligned method.

Division by duration:

Each subsequent pool is a subset of the next one: fast pool contains all configs from super_fast plus some other, medium contains all configs from fast plus some other and heavy contains all configs from medium plus some other.

To get final pool choose one frequency and duration, for example D_super_fast.

Pools can contain timestamp_column, chronos_device and timesfm_device parameters that can be filled by user. Default values are None, auto and gpu respectively.

  • Parameter timestamp_column can be used starting from fast pools.

  • Parameter chronos_device can be used starting from super_fast pools.

  • Parameter timesfm_device can be used starting from fast pools.

Note

This class requires auto extension to be installed. Read more about this at installation page.

Attributes

no_freq_super_fast

no_freq_fast

no_freq_medium

no_freq_heavy

D_super_fast

D_fast

D_medium

D_heavy

H_super_fast

H_fast

H_medium

H_heavy

MS_super_fast

MS_fast

MS_medium

MS_heavy

W_super_fast

W_fast

W_medium

W_heavy