tdfs4ds.agent.values_skill¶
VALUES skill: the values an entity takes, and the ends of a feature's distribution.
"Give me 5 TICKER values" and "which customers have the highest
sum_Transaction_Amount?" used to reach SUMMARY (the feature list) or an
invented answer. They are now answered by code from in-database reads
(:mod:tdfs4ds.feature_values). No rows leave Teradata beyond the ones shown.
========= ===================================================== ==================
Kind Question Read
========= ===================================================== ==================
entity "give me 5 TICKER values" entity_sample
extremes "which customers have the highest feature_extremes
lookup "feature_values_for
profile "customer 33357", "the features of TICKER NVDA" entity_profile
========= ===================================================== ==================
An entity value is recognised only when the store holds it (one typed-equality query per storage table), never guessed from the wording.
module-attribute
¶
_ENTITY_VALUES_RE = compile('\\b(values?|examples?|samples?|instances?|valeurs?|exemples?|[ée]chantillons?|(?:give|show|list)\\s+(?:me\\s+)?(?:\\d{1,3}|some|a\\s+few|the)|liste[rz]?|donne[sz]?(?:[-\\s]moi)?\\s+(?:\\d{1,3}|des|quelques|les)|montre[sz]?(?:[-\\s]moi)?\\s+(?:\\d{1,3}|des|quelques)|(?:which|what)\\b.*\\b(?:available|exist\\w*|stored|known)|quel(?:le)?s?\\b.*\\b(?:disponibles?|existe\\w*))\\b', IGNORECASE)
module-attribute
¶
_CATALOG_WORDS_RE = compile('\\b(features?|caract[ée]ristiques?|datasets?|jeux?\\s+de\\s+donn[ée]es|process(?:es|us)?|domains?|domaines?|columns?|colonnes?|definitions?|d[ée]finitions?|lineage|lign[ée]e)\\b', IGNORECASE)
module-attribute
¶
_EXTREMES_RE = compile('\\b(top|bottom|tail|highest|lowest|largest|smallest|biggest|max(?:imum)?|min(?:imum)?|outliers?|extremes?|extr[êe]mes?|anomal\\w*|unusual|atypiques?|aberrant\\w*|plus\\s+(?:grand|[ée]lev[ée]|petit|faible|bas|haut)\\w*|distribution|r[ée]partition)\\b', IGNORECASE)
module-attribute
¶
_NOT_LOOKUP_RE = compile('\\b(mean(?:s|ing)?|definitions?|define|signifi\\w*|veut\\s+dire|formula|formule|comput\\w*|calcul\\w*|lineage|lign[ée]e|depend\\w*|fresh\\w*|up\\s+to\\s+date|[àa]\\s+jour|derni[eè]re\\s+mise)\\b', IGNORECASE)
module-attribute
¶
_STOP_WORDS = {'THE', 'A', 'AN', 'OF', 'FOR', 'IN', 'ON', 'AT', 'TO', 'IS', 'ARE', 'WHAT', 'WHICH', 'WHO', 'HOW', 'AND', 'OR', 'ME', 'MY', 'GIVE', 'SHOW', 'TELL', 'ABOUT', 'VALUE', 'VALUES', 'FEATURE', 'FEATURES', 'PLEASE', 'IT', 'ITS', 'THIS', 'THAT', 'THESE', 'BY', 'WITH', 'ALL', 'LE', 'LA', 'LES', 'DE', 'DES', 'DU', 'POUR', 'EST', 'SONT', 'QUEL', 'QUELLE', 'QUELS', 'QUELLES', 'VALEUR', 'VALEURS', 'DONNE', 'MOI', 'MONTRE', 'ET', 'OU', 'UN', 'UNE', 'AU', 'AUX'}
Words of the question that could be entity values (the store decides).
The entity the question names, exactly or in plain words (one only).
The entity of the named features whose stored values include tokens.
The kind of values question asked (see the module docstring), or None.
Read from the store of the turn: names and values are never guessed.
The choice a reply makes among the offered (domain, entity) pairs: a
number, or words naming the entity / the domain. None when it names none.
skill_values(question: str, object_name: Optional[str] = None, entity_name: Optional[str] = None, data_domain: Optional[str] = None, as_of=None, pending: Optional[dict] = None) -> dict
Answer a VALUES question (direct answer, written by code). pending is the question this skill asked last turn ("which one?"): the reply chooses.
The question speaks of the entity in plain words: every key column, without
an id / key / code suffix and as a word stem ("customers" for
CustomerID, "categories" for Category).
{"entity", "domain"} for the one stored feature the turn means, or
{"choices": [rows]} when it stays ambiguous (never guessed).
The extremes of a feature as Markdown lines (also appended to a distribution chart).
Values of named features for given entity values: one table.
Every feature of given entity values: one table per value.