Parabon ® Crush ™ HarnessingThePowerof Extreme-ScaleComputationonDemand ® ...
-
Upload
stopspyingonme -
Category
Documents
-
view
216 -
download
0
Transcript of Parabon ® Crush ™ HarnessingThePowerof Extreme-ScaleComputationonDemand ® ...
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
1/25
2008 Parabon Inc. All rights reserved. | 1
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Harnessing The Power of
Extreme-Scale Computation on Demand
To Perform Statistical Data Mining
Parabon Crush
Microsoft Excel
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
2/25
2010 Parabon Computation, Inc. All rights reserved.
Frontier Applications: Parabon Crush
DesktopsDesktops
Servers &Servers & VMsVMs
WorkstationsWorkstations
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Microsoft Excel
Parabon Crush performs statistical data
mining and exhaustive regression analysis
from within Microsoft Excel by exercising
Frontier Grid Services.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
3/25
2008 Parabon Inc. All rights reserved. | 3
2010 Parabon Computation, Inc. All rights reserved.
Parabon CrushStatistical Data Mining at Scale
2010 Parabon Computation, Inc. All rights reserved.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
4/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Parabon Crush is a statistical data mining
application that uses the power of Frontier to
identify explanatory regression models
among the potentially vast set of all possible
models.
Unlike traditional statistical modeling tools,
which use simple heuristics to rapidly
produce answers that are often suboptimal,
Crush systematically exhausts the entirespace of possible models in its search for the
best one and it does so quickly, thanks to the
power of the Frontier Grid Platform.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
5/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
(-1)-1.5(-1)20.14.039999
424.64252.110.431003
9.810.83.141523.74.741002
2.712.11926.25.251001
3.141.3620.14.031000
XnX3X2X1Y-ValueSubject
y = x
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
6/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
(-1)-1.5(-1)20.14.039999
424.64252.110.431003
9.810.83.141523.74.741002
2.712.11926.25.251001
3.141.3620.14.031000
XnX3X2X1Y-ValueSubject
y = (1/5)x
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
7/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
(-1)-1.5(-1)20.14.039999
424.64252.110.431003
9.810.83.141523.74.741002
2.712.11926.25.251001
3.141.3620.14.031000
XnX3X2X1Y-ValueSubject
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
8/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
(-1)-1.5(-1)20.14.039999
424.64252.110.431003
9.810.83.141523.74.741002
2.712.11926.25.251001
3.141.3620.14.031000
XnX3X2X1Y-ValueSubject
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
9/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
70.3-1.5(-1)20.14.039999
12.34.64252.110.431003
31.40.83.141523.74.741002
10.52.11926.25.251001
14.31.3620.14.031000
XnX3X2X1Y-ValueSubject
y = (2)x3+(1/10)xn
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
10/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
It can be used for deep correlation analysis,
prediction and explanatory modeling.
Users have applied Crush to many domainsincluding: cancer research,
epidemiology,
nancial forecasting and
social network analysis.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
11/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Crush runs as a Microsoft Excel Add-in.
Once installed, Crush is available from the
Tools menu in Excel.
For exceptionally large datasets, Crush can
instead be launched directly from the
command-line, circumventing Excels data
size limitations.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
12/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Parameters governing a search job are
specied via standard dialog interfaces.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
13/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Crush supports linear, logit and ordered
regression, and can be extended to
handle other types of models as well.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
14/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
By default, Crush employs an exhaustive
combinatorial search. This ensures it nds the
best possible model. However, the number of
possible models is exponential in the number
of variables, so even with grid-scale power,
the time required to exhaust model spaces
with more than 40 variables is prohibitive.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
15/25
2010 Parabon Computation, Inc. All rights reserved.
k=0
n
( )nk
Parabon Crush
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
16/25
2010 Parabon Computation, Inc. All rights reserved.
k=0
n
( )nk
k=0
n
n!
k!(n-k)!
Parabon Crush
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
17/25
2010 Parabon Computation, Inc. All rights reserved.
k=0
n
n!
k!(n-k)!
Parabon Crush
Solution space grows
EXPONENTIALLY (2n)
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
18/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Solution space grows
EXPONENTIALLY (2n)
Columns25 50 100 500 1000
Years
7B
60M500K
4M30K
200
1
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
19/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
For these cases, Crush employs a sophisticated
evolutionary search algorithm to nd the best
model in the time allotted.
The ability to eectively search such large modelspaces is a breakthrough capability not found in
modeling packages that lack Frontier power.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
20/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Job output is directed to a new
workbook or new spreadsheet.
Once a job is fully specied, it can
be launched against a Frontier grid
without leaving Excel.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
21/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
After a job is launched, a placeholder for
results is created, where they are displayed
automatically when the job is complete.
The workbook can be closed and reopened atany time without aecting running jobs.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
22/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
As for all Frontier jobs, progress can
be monitored from any browser via
the Frontier Dashboard.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
23/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Jobs that would take years to complete on a
single computer can be completed in hours
or minutes, depending upon the level of grid
capacity used.
Upon completion, parameter estimates for
the best models are returned and displayed in
the workbook for further analysis.
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
24/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Computational Work Units
2010 Parabon Computation, Inc. All rights reserved.
Together, the Frontier Grid Platform and
Parabon Crush deliver a revolutionary new
combination of statistical data mining tools and
grid-scale computational power to answer deep
and valuable questions about your data - fast!
-
7/27/2019 Parabon Crush HarnessingThePowerof Extreme-ScaleComputationonDemand ToPerformStatisticalDataMining
25/25
2010 Parabon Computation, Inc. All rights reserved.
Parabon Crush
Want to learn more? Contact us at:
2010 Parabon Computation, Inc. All rights reserved.