factor_agent#
LLM-driven, mechanism-first formulaic factor research on quantlab.
Modules
Acceptance: the accepted factors' incremental test in the acceptance model, the one-time hold-out, and their export. |
|
Accepted factors exported by |
|
Session aggregations: the operators that reduce each session's intraday bars to one value per stock and day. |
|
The Campaign: the research loop, from one config to accepted and rejected candidates. |
|
A factor candidate as the proposer writes it: a stated mechanism and the KunQuant code. |
|
The |
|
The campaign config: one YAML file holds every setting of a run. |
|
Derived columns: columns a daily factor's graph reads that its dataset does not hold. |
|
The Evaluator: the only module that reads data. |
|
The events file: what a campaign is doing, appended as it happens, for anyone watching the run. |
|
Accepted factors as quantlab factors: a |
|
The Gate: deterministic checks that every candidate passes before it is evaluated. |
|
The trial ledger: one JSON line per candidate, accepted or not. |
|
Period lags read from data: the lag columns |
|
The report of |
|
Prompt templates. |
|
The Researcher: the LLM roles of a round, from prompt to checked candidates. |
|
Scoring: how a factor's raw values become the values that are judged. |
|
Evaluation segments: the date ranges the campaign and the acceptance read, the purge and the hold-out lock. |
|
Session columns: intraday graphs computed on intraday bars, one value per stock and session. |
|
The RunSnapshot: the files a run searched with and the code it ran, kept in its run directory. |
|
The web frontend: a read-only view of the run directories under a folder, served on localhost. |
|
The Zoo: the factors a candidate must not rewrite, as graphs the Gate compares it with. |