Optimize for accuracy for MTS comparison

[tallamjr]

Fawaz et al compares accuracy results instead of loss, this should then be optimised for, for a comparison

[tallamjr]

The difference seemed minimal when optimising for accuracy:

diff --git a/results/mts-t2-results.csv b/results/mts-t2-results.csv
index 841a347..ed5a848 100644
--- a/results/mts-t2-results.csv
+++ b/results/mts-t2-results.csv
@@ -1,13 +1,13 @@
 ,t2
-ArabicDigits,0.9731818437576294
+ArabicDigits,0.9750000238418579
 AUSLAN,0.929122805595398
 CharacterTrajectories,0.9456606507301331
 CMUsubject16,1.0
 ECG,0.8399999737739563
-JapaneseVowels,0.9729729890823364
+JapaneseVowels,0.9756757020950317
 KickvsPunch,0.8999999761581421
 Libras,0.8166666626930237
 NetFlow,0.779026210308075
 UWave,0.8452547788619995
 Wafer,0.8939732313156128
 WalkvsRun,1.0

May hold this branch without merging, but keep for reference.

Above command run: $ git diff -U1000 master.. -- results/mts-t2-results.csv

[tallamjr]

Close with results not changing significantly