quantlab.dataset.config.BadPrintMaskedDatasetConfig#

class quantlab.dataset.config.BadPrintMaskedDatasetConfig(*, datasets, name=None, jump=5.0, volume_ratio=20.0, lookback=20, price_variable='adjClose', volume_variable='adjVolume', masked_variables=('open', 'high', 'low', 'close', 'adjOpen', 'adjHigh', 'adjLow', 'adjClose', 'marketcap'), intrabar_variables=('adjOpen',))[source]#

Bases: MergedDatasetConfig

Config of a merged dataset whose bad prints are masked (#223).

The inputs as for MergedDatasetConfig, plus the rule of quantlab.dataset._support.cleaning.bad_print_mask and the variables it reads and masks, all in the merged panel’s shared names. The defaults name the Sharadar panels’ variables (SEP’s adjClose and adjVolume, DAILY’s marketcap); name another input’s own, or the panel raises KeyError for a variable the rule reads and none holds.

Examples

>>> cfg = BadPrintMaskedDatasetConfig(datasets=[sep, daily])
>>> cfg.jump, cfg.volume_ratio, cfg.lookback
(5.0, 20.0, 20)

intrabar_variables = ('adjOpen',)#: Adjusted intrabar prices checked on their own against the bar’s close and the close before (intrabar_bad_print_mask); a flagged one is set to NaN with its unadjusted twin (adjOpen with open) and the rest of the bar is kept. () switches the check off. The open only by default: a real day’s high or low can be far from both closes (TPST on 2023-10-11 gapped up 7 times and closed 40 times up, so its low is more than 5 times from either), while an open far from both is wrong in every case found since 2016 (25 bars, among them SLND’s 2000.00).

jump = 5.0#: Fold move beyond which a bar is a candidate (bad_print_mask).

lookback = 20#: Bars before a bar the rule reads.

masked_variables = ('open', 'high', 'low', 'close', 'adjOpen', 'adjHigh', 'adjLow', 'adjClose', 'marketcap')#: Variables set to NaN at a bad print, where the panel holds them: the prices and the market cap, which is a price times shares.

price_variable = 'adjClose'#: The adjusted price the rule reads.

volume_ratio = 20.0#: A candidate on less than this many times its lookback’s mean volume is a bad print.

volume_variable = 'adjVolume'#: The volume the rule reads, on the adjusted price’s share basis.