{-| Module : Gargantext.Database.Metrics Description : Get Metrics from Storage (Database like) Copyright : (c) CNRS, 2017-Present License : AGPL + CECILL v3 Maintainer : team@gargantext.org Stability : experimental Portability : POSIX Node API -} {-# LANGUAGE QuasiQuotes #-} module Gargantext.Database.Action.Metrics where import Database.PostgreSQL.Simple.SqlQQ (sql) import Data.HashMap.Strict (HashMap) import Data.Map (Map) import Data.Set (Set) import Database.PostgreSQL.Simple (Query, Only(..)) import Database.PostgreSQL.Simple.Types (Values(..), QualifiedIdentifier(..)) import Data.Vector (Vector) import Gargantext.Core (HasDBid(toDBid)) import Gargantext.API.Ngrams.Tools (filterListWithRoot, groupNodesByNgrams, Diagonal(..), getCoocByNgrams, mapTermListRoot, RootTerm, getRepo) import Gargantext.Database.Prelude (runPGSQuery{-, formatPGSQuery-}) import Gargantext.API.Ngrams.Types (TabType(..), ngramsTypeFromTabType, NgramsTerm(..)) import Gargantext.Core.Mail.Types (HasMail) import Gargantext.Core.NodeStory import Gargantext.Core.Text.Metrics (scored, Scored(..), {-localMetrics, toScored-}) import Database.PostgreSQL.Simple.ToField (toField, Action{-, ToField-}) import Gargantext.Core.Types (ListType(..), Limit, NodeType(..), ContextId) import Gargantext.Database.Action.Flow.Types (FlowCmdM) import Gargantext.Database.Action.Metrics.NgramsByContext (getContextsByNgramsOnlyUser{-, getTficfWith-}) import Gargantext.Database.Admin.Config (userMaster) import Gargantext.Database.Admin.Types.Node (ListId, CorpusId) import Gargantext.Database.Query.Table.Node (defaultList) import Gargantext.Database.Query.Table.Node.Select import Gargantext.Prelude import qualified Data.HashMap.Strict as HM import qualified Data.Map as Map import qualified Data.Set as Set import qualified Data.List as List import qualified Data.Text as Text getMetrics :: FlowCmdM env err m => CorpusId -> Maybe ListId -> TabType -> Maybe Limit -> m (HashMap NgramsTerm (ListType, Maybe NgramsTerm), Vector (Scored NgramsTerm)) getMetrics cId maybeListId tabType maybeLimit = do (ngs, _, myCooc) <- getNgramsCooc cId maybeListId tabType maybeLimit -- TODO HashMap pure (ngs, scored myCooc) getNgramsCooc :: (FlowCmdM env err m) => CorpusId -> Maybe ListId -> TabType -> Maybe Limit -> m ( HashMap NgramsTerm (ListType, Maybe NgramsTerm) , HashMap NgramsTerm (Maybe RootTerm) , HashMap (NgramsTerm, NgramsTerm) Int ) getNgramsCooc cId maybeListId tabType maybeLimit = do lId <- case maybeListId of Nothing -> defaultList cId Just lId' -> pure lId' (ngs', ngs) <- getNgrams lId tabType lIds <- selectNodesWithUsername NodeList userMaster myCooc <- HM.filter (>1) <$> getCoocByNgrams (Diagonal True) <$> groupNodesByNgrams ngs <$> getContextsByNgramsOnlyUser cId (lIds <> [lId]) (ngramsTypeFromTabType tabType) (take' maybeLimit $ HM.keys ngs) pure $ (ngs', ngs, myCooc) ------------------------------------------------------------------------ ------------------------------------------------------------------------ updateNgramsOccurrences :: (FlowCmdM env err m) => CorpusId -> Maybe ListId -> m () updateNgramsOccurrences cId mlId = do _ <- mapM (updateNgramsOccurrences' cId mlId Nothing) [Terms, Sources, Authors, Institutes] pure () updateNgramsOccurrences' :: (FlowCmdM env err m) => CorpusId -> Maybe ListId -> Maybe Limit -> TabType -> m [Int] updateNgramsOccurrences' cId maybeListId maybeLimit tabType = do lId <- case maybeListId of Nothing -> defaultList cId Just lId' -> pure lId' result <- getNgramsOccurrences cId lId tabType maybeLimit let toInsert :: [[Action]] toInsert = map (\(ngramsTerm, score) -> [ toField cId , toField lId , toField $ unNgramsTerm ngramsTerm , toField $ toDBid $ ngramsTypeFromTabType tabType , toField score ] ) $ HM.toList result queryInsert :: Query queryInsert = [sql| WITH input(corpus_id, list_id, terms, type_id, weight) AS (?) INSERT into node_node_ngrams (node1_id, node2_id, ngrams_id, ngrams_type, weight) SELECT input.corpus_id,input.list_id,ngrams.id,input.type_id,input.weight FROM input JOIN ngrams on ngrams.terms = input.terms ON CONFLICT (node1_id, node2_id, ngrams_id, ngrams_type) DO UPDATE SET weight = excluded.weight RETURNING 1 |] let fields = map (\t-> QualifiedIdentifier Nothing t) $ map Text.pack ["int4", "int4","text","int4","int4"] map (\(Only a) -> a) <$> runPGSQuery queryInsert (Only $ Values fields toInsert) ------------------------------------------------------------------------ -- Used for scores in Ngrams Table getNgramsOccurrences :: (FlowCmdM env err m) => CorpusId -> ListId -> TabType -> Maybe Limit -> m (HashMap NgramsTerm Int) getNgramsOccurrences c l t ml = HM.map Set.size <$> getNgramsContexts c l t ml getNgramsContexts :: (FlowCmdM env err m) => CorpusId -> ListId -> TabType -> Maybe Limit -> m (HashMap NgramsTerm (Set ContextId)) getNgramsContexts cId lId tabType maybeLimit = do (_ngs', ngs) <- getNgrams lId tabType lIds <- selectNodesWithUsername NodeList userMaster -- TODO maybe add an option to group here getContextsByNgramsOnlyUser cId (lIds <> [lId]) (ngramsTypeFromTabType tabType) (take' maybeLimit $ HM.keys ngs) ------------------------------------------------------------------------ updateContextScore :: (FlowCmdM env err m) => CorpusId -> Maybe ListId -> m [Int] updateContextScore cId maybeListId = do lId <- case maybeListId of Nothing -> defaultList cId Just lId' -> pure lId' result <- getContextsNgramsScore cId lId Terms MapTerm Nothing let toInsert :: [[Action]] toInsert = map (\(contextId, score) -> [ toField cId , toField contextId , toField score ] ) $ Map.toList result queryInsert :: Query queryInsert = [sql| WITH input(node_id, context_id, score) AS (?) UPDATE nodes_contexts nc SET score = input.score FROM input WHERE nc.node_id = input.node_id AND nc.context_id = input.context_id RETURNING 1 |] let fields = map (\t-> QualifiedIdentifier Nothing t) $ map Text.pack ["int4", "int4","int4"] map (\(Only a) -> a) <$> runPGSQuery queryInsert (Only $ Values fields toInsert) -- Used for scores in Doc Table getContextsNgramsScore :: (FlowCmdM env err m) => CorpusId -> ListId -> TabType -> ListType -> Maybe Limit -> m (Map ContextId Int) getContextsNgramsScore cId lId tabType listType maybeLimit = Map.map Set.size <$> getContextsNgrams cId lId tabType listType maybeLimit getContextsNgrams :: (FlowCmdM env err m) => CorpusId -> ListId -> TabType -> ListType -> Maybe Limit -> m (Map ContextId (Set NgramsTerm)) getContextsNgrams cId lId tabType listType maybeLimit = do (ngs', ngs) <- getNgrams lId tabType lIds <- selectNodesWithUsername NodeList userMaster result <- groupNodesByNgrams ngs <$> getContextsByNgramsOnlyUser cId (lIds <> [lId]) (ngramsTypeFromTabType tabType) ( take' maybeLimit $ HM.keys $ HM.filter (\v -> fst v == listType) ngs' ) -- printDebug "getCoocByNgrams" result pure $ Map.fromListWith (<>) $ List.concat $ map (\(ng, contexts) -> List.zip (Set.toList contexts) (List.cycle [Set.singleton ng])) $ HM.toList result ------------------------------------------------------------------------ ------------------------------------------------------------------------ getNgrams :: (HasMail env, HasNodeStory env err m) => ListId -> TabType -> m ( HashMap NgramsTerm (ListType, Maybe NgramsTerm) , HashMap NgramsTerm (Maybe RootTerm) ) getNgrams lId tabType = do lists <- mapTermListRoot [lId] (ngramsTypeFromTabType tabType) <$> getRepo [lId] -- TODO filterListWithRoot [MapTerm, StopTerm, CandidateTerm] lists let maybeSyn = HM.unions $ map (\t -> filterListWithRoot t lists) [[MapTerm], [StopTerm], [CandidateTerm]] pure (lists, maybeSyn) -- Some useful Tools take' :: Maybe Int -> [a] -> [a] take' Nothing xs = xs take' (Just n) xs = take n xs