Fuzzy rule based system for anomalies detection
cabal init -n -l BSD3 --is-executable --language=Haskell2010 -u smunix.com -a 'Providence Salumu' -c Numeric -s 'Writing fuzzy rules based system' -p fuzzy
A toy fuzzy rule based system I wrote to play with anomalies detections on exchanges. Below are steps to run it:
╭─smunix at smunix in ~/Programming/haskell/fuzzy on master✘✘✘ using ‹› 15-09-14 - 2:51:31 ╰─○ cabal configure cabal configure Resolving dependencies... Configuring fuzzy-0.1.0.0...
╭─smunix at smunix in ~/Programming/haskell/fuzzy on master✘✘✘ using ‹› 15-09-14 - 2:52:36 ╰─○ cabal build cabal build Building fuzzy-0.1.0.0... Preprocessing library fuzzy-0.1.0.0... [1 of 1] Compiling Fuzzy ( src/Fuzzy.hs, dist/build/Fuzzy.o ) In-place registering fuzzy-0.1.0.0... Preprocessing executable 'fuzzy' for fuzzy-0.1.0.0... [1 of 2] Compiling Fuzzy ( src/Fuzzy.hs, dist/build/fuzzy/fuzzy-tmp/Fuzzy.o ) [2 of 2] Compiling Main ( src/Main.hs, dist/build/fuzzy/fuzzy-tmp/Main.o ) Linking dist/build/fuzzy/fuzzy ...
In the output below, you can see different alert levels we get for different notional values on both the exchanges and the firm. ╭─smunix at smunix in ~/Programming/haskell/fuzzy on master✘✘✘ using ‹› 15-09-14 - 2:52:40 ╰─○ dist/build/fuzzy/fuzzy dist/build/fuzzy/fuzzy exec {ms=Var 100.0, mkt=Var 200.0} ==> Alert=43.13838185139064% exec {ms=Var 100.0, mkt=Var 400.0} ==> Alert=43.13838185139064% exec {ms=Var 100.0, mkt=Var 600.0} ==> Alert=43.13838185139064% exec {ms=Var 100.0, mkt=Var 800.0} ==> Alert=43.13838185139064% exec {ms=Var 100.0, mkt=Var 200.0} ==> Alert=43.13838185139064% exec {ms=Var 300.0, mkt=Var 200.0} ==> Alert=43.13984905285267% exec {ms=Var 500.0, mkt=Var 200.0} ==> Alert=50.123233797386874% exec {ms=Var 700.0, mkt=Var 200.0} ==> Alert=94.06152030139934% exec {ms=Var 900.0, mkt=Var 200.0} ==> Alert=94.2655776014311% exec {ms=Var 1000.0, mkt=Var 200.0} ==> Alert=94.27189264293638% exec {ms=Var 100.0, mkt=Var 200.0} ==> Alert=43.13838185139064% exec {ms=Var 300.0, mkt=Var 200.0} ==> Alert=43.13984905285267% exec {ms=Var 500.0, mkt=Var 400.0} ==> Alert=50.247414366516935% exec {ms=Var 700.0, mkt=Var 700.0} ==> Alert=43.221484622722556% exec {ms=Var 900.0, mkt=Var 800.0} ==> Alert=43.1447221653757% exec {ms=Var 1000.0, mkt=Var 950.0} ==> Alert=43.1383462743783%
╭─smunix at smunix in ~/Programming/haskell/fuzzy on master✘✘✘ using ‹› 15-09-14 - 2:52:58 ╰─○ You can define as many rules for alerts as you see fit, and fix the tresholds and resolution as you like. rule1 ms mkt = (ms =? ms_nv_low) &? (mkt =? mkt_nv_low) ==> very alert_low rule2 ms mkt = (ms =? ms_nv_low) &? (mkt =? mkt_nv_high) ==> very alert_low rule3 ms mkt = (ms =? ms_nv_high) &? (mkt =? mkt_nv_low) ==> very alert_high rule4 ms mkt = (ms =? ms_nv_high) &? (mkt =? mkt_nv_medium) ==> alert_high rule5 ms mkt = (ms =? ms_nv_medium) &? (mkt =? mkt_nv_medium) ==> very alert_low rule6 ms mkt = (ms =? ms_nv_high) &? (mkt =? mkt_nv_high) ==> very alert_low
The only deffuzifier I have implememted by now is a Centroid. alertCentroid :: Centroid (Var Double) alertCentroid = Centroid { domainMin = Var 0, domainMax = Var 100, resolution = Var 0.5 }
But several functions can be used to define linguistic variables in the domain space:
- up : left shoulder function
- down : right shoulder function
- sigmoid
- gauss : hat like /gaussian function
- triangle (todo)