Sunday and Monday => close
Tuesday, Wednesday, Friday => till 19:00
Thursday   => till 18:00
Saturday  till 17:00

Figure of NN with 4 input 1 output

diwsplay examples of data

normal => (0.3 0.5 0.2 0.4)



<<0>>
[One is normal while the others abnormal]

Assuming one out of three families of iris flower to represent illegal
transactions while the remaining two families represent legal ones. 
Is it possible then to simulate a system for network intrusion detection
by using part of this dataset to train and remaining data to test the system?


(0)
------------------------
Train and test with
One family => attack
the other two => normal
------------------------



<<1>>
[Mutants of normal and attack]
(1) Train the system, in the same way as Problem 1, 
using three families of iris flower as normal and abnormal.
(2) All points of all the three families of iris flower 
are given mutation. Call them mutants-of-normal and mutants-of-attack.
(3) Then the system trained in step 1 can recognize these
mutants properly as normal or attack?



(1)
------------------------
Train with
One family => attack
the other two => normal

test with 
mutant of attack and normal
------------------------





[Randomly located normals and attacks]
(1) Create 50 normal samples and 100 abnormal samples all at random. 
(2) Then train the system using all of these samples of normal and attack.
(3) Test the system again with these samples of normal and attack.


(2)
----------------------------------------------------------------
Create 50 normal samples and 100 abnormal samples all at random. 
Train with half of normal and atacks
test with remaining half of the normal and attacks
----------------------------------------------------------------



[random discretely located normal and attack samples]
A set of normal and a set of abnormal samples are specified at random one by one.
Then can the system trained by those samples recognize a mutant of abnormal sample?


(3)
----------------------------------------------------------------
Create 50 normal samples and 100 abnormal samples all at random. 
Train with half of normal and atacks
test with mutant of the remaining half of the normal and attacks
----------------------------------------------------------------




[Can a sommelier be trained without bootlegs?]
[Like a wine tasting]

(1) Assume one family of iris as normal while the other two abnormal.
(2) Furthermore, randomly create an attack dataset. Call them dummy attacks.
(3) Train your intrusion detection system only with the normal set.
(4) Then, try two tests, one with only abnormal, and 
the other with only dummy, avoiding any a~priori prediction.



(4)
------------------------------------------------
assume again
one family is normal while the other two attack

train with
one family as normal

test-1 with the other family as attack
test-2 with the randomly specified data as attack
-------------------------------------------------


(5)
[A placebo experiment]
done