quantlab.dataset.base#
Dataset base classes: the read interface every dataset answers.
This module is the first layer of the pipeline (data, then factors, models
and backtests). Every layer exchanges data as a panel: an
xarray.Dataset indexed by timestamp and symbol, with one data
variable per field (open, close, volume and so on). Panels are
stored on disk as Zarr directories (a chunked array format), called stores
below.
BaseDataset is the read interface shared by every dataset: a config with
normalised ISO dates, panel(start, end, symbols, variables) (the one read
seam of every leaf dataset, recorded in an open DataRecorder),
calendar, bar_before and bar_after, resampling, the shared
variable names, and each dataset’s declaration of the stores it reads
(inputs, store_uses, stores). MarketDataset adds the market rules
(tradable and delisting bars) and the to_kunquant export (arrays for the
KunQuant factor engine).
Neither holds a build path. A dataset converted from a vendor’s raw files
into a store it owns derives from quantlab.dataset.raw.RawDataset or
RawMarketDataset and implements _raw_data_to_xr; a view (a roster
cut, a merge, a mapping) and a panel held in memory derive from the classes
here and have no build methods. The factor layer asks a dataset for a date
range with panel(start, end) and counts warm-up with
bar_before(date, n).
TickerLookup names a dataset’s symbol ids as of a day (ticker and
company); a dataset says which one applies with ticker_lookup().
Classes
The read interface shared by every dataset in quantlab. |
|
Dataset of market price bars, with a KunQuant export. |
|
The readable name of one symbol on one day: its ticker and company. |
|
Names a dataset's symbol ids as of a day: the ticker and company in use. |
Functions
Return the bar |
|
Return the bar |
|
Return |