2 Module : Gargantext.Core.Viz.Graph.Tools
3 Description : Tools to build Graph
4 Copyright : (c) CNRS, 2017-Present
5 License : AGPL + CECILL v3
6 Maintainer : team@gargantext.org
7 Stability : experimental
12 {-# LANGUAGE BangPatterns, ScopedTypeVariables #-}
14 module Gargantext.Core.Viz.Graph.Tools
18 import Data.HashMap.Strict (HashMap)
19 import Data.Map.Strict (Map)
20 import Data.Maybe (fromMaybe)
21 import Data.Swagger hiding (items)
22 import GHC.Float (sin, cos)
23 import GHC.Generics (Generic)
24 import Gargantext.API.Ngrams.Types (NgramsTerm(..))
25 import Gargantext.Core.Methods.Similarities (Similarity(..), measure)
26 -- import Gargantext.Core.Methods.Similarities.Conditional (conditional)
27 import Gargantext.Core.Statistics
28 import Gargantext.Core.Viz.Graph.Bridgeness (bridgeness, Bridgeness(..), Partitions, nodeId2comId{-, recursiveClustering-}, recursiveClustering', setNodes2clusterNodes)
29 import Gargantext.Core.Viz.Graph.Index (createIndices, toIndex, map2mat, mat2map, Index, MatrixShape(..))
30 import Gargantext.Core.Viz.Graph.Tools.IGraph (mkGraphUfromEdges, spinglass, spinglass')
31 import Gargantext.Core.Viz.Graph.Tools.Infomap (infomap)
32 import Gargantext.Core.Viz.Graph.Types (Attributes(..), Edge(..), Graph(..), MultiPartite(..), Node(..), Partite(..), Strength(..))
33 import Gargantext.Database.Schema.Ngrams (NgramsType(..))
34 import Gargantext.Core.Viz.Graph.Utils (edgesFilter, nodesFilter)
35 import Gargantext.Prelude
36 import Graph.Types (ClusterNode)
37 import IGraph.Random -- (Gen(..))
38 import Test.QuickCheck (elements)
39 import Test.QuickCheck.Arbitrary
40 import qualified Data.HashMap.Strict as HashMap
41 import qualified Data.List as List
42 import qualified Data.Map.Strict as Map
43 import qualified Data.Set as Set
44 import qualified Data.HashSet as HashSet
45 import qualified Data.Text as Text
46 import qualified Data.Vector.Storable as Vec
47 import qualified Graph.BAC.ProxemyOptim as BAC
48 import qualified IGraph as Igraph
49 import qualified IGraph.Algorithms.Layout as Layout
51 data PartitionMethod = Spinglass | Confluence | Infomap
52 deriving (Generic, Eq, Ord, Enum, Bounded, Show)
53 instance FromJSON PartitionMethod
54 instance ToJSON PartitionMethod
55 instance ToSchema PartitionMethod
56 instance Arbitrary PartitionMethod where
57 arbitrary = elements [ minBound .. maxBound ]
59 data BridgenessMethod = BridgenessMethod_Basic | BridgenessMethod_Advanced
60 deriving (Generic, Eq, Ord, Enum, Bounded, Show)
61 instance FromJSON BridgenessMethod
62 instance ToJSON BridgenessMethod
63 instance ToSchema BridgenessMethod
64 instance Arbitrary BridgenessMethod where
65 arbitrary = elements [ minBound .. maxBound ]
68 -------------------------------------------------------------
69 defaultClustering :: Map (Int, Int) Double -> IO [ClusterNode]
70 -- defaultClustering x = pure $ BAC.defaultClustering x
71 defaultClustering x = spinglass 1 x
73 -------------------------------------------------------------
74 type Threshold = Double
77 cooc2graph' :: Ord t => Similarity
80 -> Map (Index, Index) Double
81 cooc2graph' distance threshold myCooc
82 = Map.filter (> threshold)
86 Conditional -> map2mat Triangle 0 tiSize
87 Distributional -> map2mat Square 0 tiSize
88 $ Map.filter (> 1) myCooc'
91 (ti, _) = createIndices myCooc
93 myCooc' = toIndex ti myCooc
97 -- coocurrences graph computation
98 cooc2graphWith :: PartitionMethod
104 -> HashMap (NgramsTerm, NgramsTerm) Int
106 cooc2graphWith Spinglass = cooc2graphWith' (spinglass 1)
107 cooc2graphWith Confluence= cooc2graphWith' (\x -> pure $ BAC.defaultClustering x)
108 cooc2graphWith Infomap = cooc2graphWith' (infomap "-v -N2")
109 --cooc2graphWith Infomap = cooc2graphWith' (infomap "--silent --two-level -N2")
110 -- TODO: change these options, or make them configurable in UI?
112 cooc2graphWith' :: Partitions
118 -> HashMap (NgramsTerm, NgramsTerm) Int
120 cooc2graphWith' _doPartitions _bridgenessMethod multi similarity@Conditional threshold strength myCooc = do
121 let (distanceMap, diag, ti) = doSimilarityMap similarity threshold strength myCooc
122 distanceMap `seq` diag `seq` ti `seq` return ()
124 partitions <- if (Map.size distanceMap > 0)
125 -- then recursiveClustering doPartitions distanceMap
126 then recursiveClustering' (spinglass' 1) distanceMap
127 else panic $ Text.unlines [ "[Gargantext.C.V.Graph.Tools] Similarity Matrix is empty"
128 , "Maybe you should add more Map Terms in your list"
131 length partitions `seq` return ()
134 !confluence' = BAC.computeConfluences 3 (Map.keys distanceMap) True
135 !bridgeness' = bridgeness (Bridgeness_Recursive partitions 1.0 similarity) distanceMap
137 !bridgeness' = if bridgenessMethod == BridgenessMethod_Basic
138 then bridgeness (Bridgeness_Basic partitions 1.0) distanceMap
139 else bridgeness (Bridgeness_Advanced similarity confluence') distanceMap
141 pure $ data2graph multi ti diag bridgeness' confluence' (setNodes2clusterNodes $ List.concat partitions)
143 cooc2graphWith' _doPartitions _bridgenessMethod multi similarity@Distributional threshold strength myCooc = do
144 let (distanceMap, diag, ti) = doSimilarityMap Distributional threshold strength myCooc
145 distanceMap `seq` diag `seq` ti `seq` return ()
147 partitions <- if (Map.size distanceMap > 0)
148 --then recursiveClustering doPartitions distanceMap
149 then recursiveClustering' (spinglass' 1) distanceMap
150 else panic $ Text.unlines [ "[Gargantext.C.V.Graph.Tools] Similarity Matrix is empty"
151 , "Maybe you should add more Map Terms in your list"
154 length partitions `seq` return ()
157 !confluence' = BAC.computeConfluences 3 (Map.keys distanceMap) True
158 !bridgeness' = bridgeness (Bridgeness_Recursive partitions 1.0 similarity) distanceMap
160 !bridgeness' = if bridgenessMethod == BridgenessMethod_Basic
161 then bridgeness (Bridgeness_Basic partitions 1.0) distanceMap
162 else bridgeness (Bridgeness_Advanced Distributional confluence') distanceMap
163 pure $ data2graph multi ti diag bridgeness' confluence' partitions
165 pure $ data2graph multi ti diag bridgeness' confluence' (setNodes2clusterNodes $ List.concat partitions)
171 doSimilarityMap :: Similarity
174 -> HashMap (NgramsTerm, NgramsTerm) Int
175 -> ( Map (Int,Int) Double
176 , Map (Index, Index) Int
177 , Map NgramsTerm Index
180 doSimilarityMap Conditional threshold strength myCooc = (distanceMap, toIndex ti myCooc', ti)
182 myCooc' = Map.fromList $ HashMap.toList myCooc
184 (_diag, theMatrix) = Map.partitionWithKey (\(x,y) _ -> x == y)
186 $ HashMap.toList myCooc
188 (ti, _it) = createIndices theMatrix
191 similarities = (\m -> m `seq` m)
192 $ (\m -> m `seq` measure Conditional m)
193 $ (\m -> m `seq` map2mat Square 0 tiSize m)
194 $ theMatrix `seq` toIndex ti theMatrix
196 links = round (let n :: Double = fromIntegral (Map.size ti) in 10 * n * (log n)^(2::Int))
197 distanceMap = Map.fromList
199 $ (if strength == Weak then List.reverse else identity)
202 $ Map.filter (> threshold)
203 -- $ conditional myCooc
204 $ similarities `seq` mat2map similarities
206 doSimilarityMap Distributional threshold strength myCooc = (distanceMap, toIndex ti diag, ti)
209 (diag, theMatrix) = Map.partitionWithKey (\(x,y) _ -> x == y)
211 $ HashMap.toList myCooc
213 (ti, _it) = createIndices theMatrix
216 similarities = (\m -> m `seq` m)
217 $ (\m -> m `seq` measure Distributional m)
218 $ (\m -> m `seq` map2mat Square 0 tiSize m)
219 $ theMatrix `seq` toIndex ti theMatrix
221 links = round (let n :: Double = fromIntegral tiSize in n * (log n)^(2::Int))
223 distanceMap = Map.fromList
225 $ (if strength == Weak then List.reverse else identity)
229 $ (\m -> m `seq` Map.filter (> threshold) m)
230 $ similarities `seq` mat2map similarities
232 ----------------------------------------------------------
233 -- | From data to Graph
234 type Occurrences = Int
236 nodeTypeWith :: MultiPartite -> NgramsTerm -> NgramsType
237 nodeTypeWith (MultiPartite (Partite s1 t1) (Partite _s2 t2)) t =
238 if HashSet.member t s1
243 data2graph :: MultiPartite
244 -> Map NgramsTerm Int
245 -> Map (Int, Int) Occurrences
246 -> Map (Int, Int) Double
247 -> Map (Int, Int) Double
250 data2graph multi labels' occurences bridge conf partitions =
251 Graph { _graph_nodes = nodes
252 , _graph_edges = edges
253 , _graph_metadata = Nothing
258 nodes = map (setCoord ForceAtlas labels bridge)
259 [ (n, Node { node_size = maybe 0 identity (Map.lookup (n,n) occurences)
260 , node_type = nodeTypeWith multi label
261 , node_id = (cs . show) n
262 , node_label = unNgramsTerm label
266 Attributes { clust_default = fromMaybe 0
267 (Map.lookup n community_id_by_node_id)
272 | (label, n) <- labels
273 , Set.member n toKeep
276 (bridge', toKeep) = nodesFilter (\v -> v > 1) bridge
278 edges = [ Edge { edge_source = cs (show s)
279 , edge_hidden = Nothing
280 , edge_target = cs (show t)
281 , edge_weight = weight
282 , edge_confluence = maybe 0 identity $ Map.lookup (s,t) conf
283 , edge_id = cs (show i)
285 | (i, ((s,t), weight)) <- zip ([0..]::[Integer] ) $ Map.toList bridge'
290 community_id_by_node_id = Map.fromList
291 $ map nodeId2comId partitions
293 labels = Map.toList labels'
296 ------------------------------------------------------------------------
298 data Layout = KamadaKawai | ACP | ForceAtlas
301 setCoord' :: (Int -> (Double, Double)) -> (Int, Node) -> Node
302 setCoord' f (i,n) = n { node_x_coord = x, node_y_coord = y }
308 setCoord :: Ord a => Layout -> [(a, Int)] -> Map (Int, Int) Double -> (Int, Node) -> Node
309 setCoord l labels m (n,node) = node { node_x_coord = x
313 (x,y) = getCoord l labels m n
319 -> Map (Int, Int) Double
322 getCoord KamadaKawai _ _m _n = undefined -- layout m n
324 getCoord ForceAtlas _ _ n = (sin d, cos d)
328 getCoord ACP labels m n = to2d $ maybe (panic "Graph.Tools no coordinate") identity
330 $ pcaReduceTo (Dimension 2)
333 to2d :: Vec.Vector Double -> (Double, Double)
336 ds = take 2 $ Vec.toList v
340 mapArray :: Ord a => [(a, Int)] -> Map (Int, Int) Double -> Map Int (Vec.Vector Double)
341 mapArray items m' = Map.fromList [ toVec n' ns m' | n' <- ns ]
345 toVec :: Int -> [Int] -> Map (Int,Int) Double -> (Int, Vec.Vector Double)
346 toVec n' ns' m'' = (n', Vec.fromList $ map (\n'' -> maybe 0 identity $ Map.lookup (n',n'') m'') ns')
347 ------------------------------------------------------------------------
349 -- | KamadaKawai Layout
350 -- TODO TEST: check labels, nodeId and coordinates
351 layout :: Map (Int, Int) Double -> Int -> Gen -> (Double, Double)
352 layout m n gen = maybe (panic "") identity $ Map.lookup n $ coord
354 coord :: (Map Int (Double,Double))
355 coord = Map.fromList $ List.zip (Igraph.nodes g) $ (Layout.layout g p gen)
356 --p = Layout.defaultLGL
357 p = Layout.kamadaKawai
358 g = mkGraphUfromEdges $ map fst $ List.filter (\e -> snd e > 0) $ Map.toList m
360 -----------------------------------------------------------------------------
362 cooc2graph'' :: Ord t => Similarity
365 -> Map (Index, Index) Double
366 cooc2graph'' distance threshold myCooc = neighbourMap
368 (ti, _) = createIndices myCooc
369 myCooc' = toIndex ti myCooc
370 matCooc = map2mat Triangle 0 (Map.size ti) $ Map.filter (> 1) myCooc'
371 distanceMat = measure distance matCooc
372 neighbourMap = filterByNeighbours threshold
373 $ mat2map distanceMat
376 filterByNeighbours :: Double -> Map (Index, Index) Double -> Map (Index, Index) Double
377 filterByNeighbours threshold distanceMap = filteredMap
380 indexes = List.nub $ List.concat $ map (\(idx,idx') -> [idx,idx'] ) $ Map.keys distanceMap
381 filteredMap :: Map (Index, Index) Double
382 filteredMap = Map.fromList
385 let selected = List.reverse
389 $ Map.filterWithKey (\(from,_) _ -> idx == from) distanceMap
390 in List.take (round threshold) selected