Databases Reference
In-Depth Information
Algorithm 3
Visualizing concept drift from
t
m
to
t
n
1: Retrieve the compressed
TSP
trie
TSP
i
indexed by (
t
istart
,t
iend
), where
t
istart
≤ t
m
≤ t
iend
2:
Listl.push back
(
TSPi
)
3:
t
=
t
i
end
4:
while
existing a compressed
TSP
trie
TSP
k
indexed by (
t
kstart
,t
kend
), where
t ≤ t
kstart
≤ t
n
do
5:
Listl.push back
(
TSP
k
)
6:
t
=
t
kend
7:
end while
8:
while
l
is not empty
do
9:
TSP tsp
=
l.pop front
()
10: Retrieve and visualize clustering pattern from
tsp
11: time-delay()
12:
end while
6.4 Visualizing Micro-shift with a Concept Cycle
We can not only retrieval and visualize cluster pattern for each concept cycle,
but also zoom into each active cell to detect the micro-shift of that cell over
the concept cycle. Algorithm 4 shows the steps for this purpose.
Besides the above patterns, we can also retrieve attribute correlation pat-
terns, and clustering patterns in certain subspace for a concept cycle, as well
as the drifts of the above patterns over multiple cycles based on compressed
TSP tries. We will report these algorithms in a separate paper.
Algorithm 4
Visualizing the micro-shift of an active cell
cell
i
within a con-
cept cycle around
t
x
1: Retrieve the compressed TSP trie indexed by (
t
start
,t
end
), where
t
start
≤ t
x
≤
t
end
.
2: Specify the visualization resolution
n
, based on which generate a list of
timestamps
l
t
=(
t
start
,t
2
, ··· ,t
n−
2
,t
end
)
3:
while
l
t
is not empty
do
4:
t
=
l
t
.pop front
()
5:
Calculate the number of points, mean, and standard deviation of
cell
i
at
t
by
using the following formulas
• n
t
=
a
n
t
x
+
b
n
• m
t
=
a
m
t
x
+
b
m
• d
t
=
a
d
t
x
+
b
d
6: Then use
n
i
,m
i
,d
i
to deploy data in
cell
i
at the n-dimensional space
7: time-delay()
8:
end while