SlideShare a Scribd company logo
1
Adjusting PageRank parameters and
Comparing results
Subhajit Sahu1
, Kishore Kothapalli1
, Dip Sankar Banerjee2
1
International Institute of Information Technology, Hyderabad
2
Indian Institute of Technology, Jodhpur
Abstract — The effect of adjusting damping factor α and tolerance τ on iterations
needed for PageRank computation is studied here. Relative performance of
PageRank computation with L1, L2, and L∞ norms used as convergence check, are
also compared with six possible mean ratios. It is observed that increasing the
damping factor α linearly increases the iterations needed almost exponentially. On
the other hand, decreasing the tolerance τ exponentially decreases the iterations
needed almost exponentially. On average, PageRank with L∞ norm as convergence
check is the fastest, quickly followed by L2 norm, and then L1 norm. For large
graphs, above certain tolerance τ values, convergence can occur in a single iteration.
On the contrary, below certain tolerance τ values, sensitivity issues can begin to
appear, causing computation to halt at maximum iteration limit without convergence.
The six mean ratios for relative performance comparison are based on arithmetic,
geometric, and harmonic mean, as well as the order of ratio calculation. Among
them GM-RATIO, geometric mean followed by ratio calculation, is found to be most
stable, followed by AM-RATIO.
Index terms — PageRank algorithm, Parameter adjustment, Convergence function,
Sensitivity issues, Relative performance comparison.
1. Introduction
Web graphs unaltered are reducible, and thus the rate of convergence of the
power-iteration method is the rate at which αk
→ 0, where α is the damping
factor, and k is the iteration count. An estimate of the number of iterations
needed to converge to a tolerance τ is log10 τ / log10 α [1]. For τ = 10-6
and α =
0.85, it can take roughly 85 iterations to converge. For α = 0.95, and α = 0.75,
with the same tolerance τ = 10-6
, it takes roughly 269 and 48 iterations
respectively. For τ = 10-9
, and τ = 10-3
, with the same damping factor α =
2
0.85, it takes roughly 128 and 43 iterations respectively. Thus, adjusting the
damping factor or the tolerance parameters of the PageRank algorithm can
have a significant effect on the convergence rate.
However, once results for various test cases are obtained, there exist
multiple methods to obtain a composite relative performance ratio.
Consider, for example, three approaches a, b, and c, with 3 test runs for each
of the three approaches, labeled a1, a2, a3, b1, b2, b3, c1, c2, c3. One method to
get a composite ratio between the three approaches would be to find the
relative performance ratio of each approach with respect to a baseline
approach (one of them), and then calculate the arithmetic-mean (AM) for
each approach. For example, the relative performance of each approach with
respect to c would be a1/c1, b1/c1, c1/c1, a2/c2, b2/c2, and so on. The RATIO-AM
with respect to c is now the arithmetic mean of these ratios, i.e.,
(a1/c1+a2/c2+a3/c3)/3 for a, (b1/c1+b2/c2+b3/c3)/3, and 1 for c. Similarly,
RATIO-GM, and RATIO-HM can be obtained by instead calculating geometric
mean (GM), or harmonic mean (HM) respectively. Unfortunately, based upon
the choice of the baseline approach, the composite ratios can differ (except
RATIO-GM, as discussed later). The alternative approach is to calculate
means for each approach first, and then find the relative performance ratio.
Like before, arithmetic, geometric, or harmonic mean can be used, called
AM-RATIO, GM-RATIO, and HM-RATIO respectively. For example, the
arithmetic mean of each approach would be (a1+a2+a3)/3 for a, (b1+b2+b3)/3
for b, and so on. The AM-RATIO with respect to c is now the ratio of means
of each approach with respect to c’s, i.e., (a1+a2+a3)/(c1+c2+c3) for a,
(b1+b2+b3)/(c1+c2+c3) for b, and 1 for c. Since the ratio is calculated last, the
choice of the baseline approach has no effect on the result. Note that,
GM-RATIO and RATIO-GM give identical composite ratios due to the
mathematical property of geometric mean, and thus the order of ratio
calculation does not affect the result. Weighted geometric mean has been
selected for SPECviewperf® composite numbers [2].
2. Adjusting Damping factor
Adjustment of the damping factor α is a delicate balancing act. For smaller
values of α, the convergence is fast, but the link structure of the graph used
to determine ranks is less true. Slightly different values for α can produce
3
very different rank vectors. Moreover, as α → 1, convergence slows down
drastically, and sensitivity issues begin to surface [1].
For this experiment, the damping factor α (which is usually 0.85) is varied
from 0.50 to 1.00 in steps of 0.05. This is in order to compare the
performance variation with each damping factor. The calculated error is the
L1 norm with respect to default PageRank (α = 0.85). The PageRank
algorithm used here is the standard power-iteration (pull) based PageRank.
The rank of a vertex in an iteration is calculated as c0 + αΣrn/dn, where c0 is
the common teleport contribution, α is the damping factor, rn is the previous
rank of vertex with an incoming edge, dn is the out-degree of the
incoming-edge vertex, and N is the total number of vertices in the graph. The
common teleport contribution c0, calculated as (1-α)/N + αΣrn/N, includes
the contribution due to a teleport from any vertex in the graph due to the
damping factor (1-α)/N, and teleport from dangling vertices (with no
outgoing edges) in the graph αΣrn/N. This is because a random surfer jumps
to a random page upon visiting a page with no links, in order to avoid the
rank-sink effect.
All seventeen graphs used in this experiment are stored in the MatrixMarket
(.mtx) file format, and obtained from the SuiteSparse Matrix Collection.
These include: web-Stanford, web-BerkStan, web-Google, web-NotreDame,
soc-Slashdot0811, soc-Slashdot0902, soc-Epinions1, coAuthorsDBLP,
coAuthorsCiteseer, soc-LiveJournal1, coPapersCiteseer, coPapersDBLP,
indochina-2004, italy_osm, great-britain_osm, germany_osm, asia_osm. The
experiment is implemented in C++, and compiled using GCC 9 with
optimization level 3 (-O3). The system used is a Dell PowerEdge R740 Rack
server with two Intel Xeon Silver 4116 CPUs @ 2.10GHz, 128GB DIMM
DDR4 Synchronous Registered (Buffered) 2666 MHz (8x16GB) DRAM, and
running CentOS Linux release 7.9.2009 (Core). The iterations taken with
each test case is measured. 500 is the maximum iterations allowed.
Statistics of each test case is printed to standard output (stdout), and
redirected to a log file, which is then processed with a script to generate a
CSV file, with each row representing the details of a single test case. This
CSV file is imported into Google Sheets, and necessary tables are set up
with the help of the FILTER function to create the charts.
4
Figure 2.1: Geometric mean iterations for PageRank computation with damping
factor α adjusted from 0.50 - 1.00 in steps of 0.05. Chart for AM iterations is quite
similar.
Figure 2.2: Relative geometric mean (GM-RATIO) iterations for PageRank
computation with damping factor α adjusted from 0.50 - 1.00 in steps of 0.05. Chart
for AM-RATIO is quite similar.
Results, as shown in figures 2.1 and 2.2, indicate that increasing the
damping factor α beyond 0.85 significantly increases convergence time,
and lowering it below 0.85 decreases convergence time. As the damping
factor α increases linearly, the iterations needed for PageRank computation
5
increases almost exponentially. On average, using a damping factor α = 0.95
increases iterations needed by 190% (~2.9x), and using a damping factor α =
0.75 decreases it by 41% (~0.6x), compared to damping factor α = 0.85.
Note that a higher damping factor implies that a random surfer follows links
with higher probability (and jumps to a random page with lower probability).
3. Adjusting Error function
It is observed that a number of error functions are in use for checking
convergence of PageRank computation. Although L1 norm is commonly used
for convergence check, it appears nvGraph uses L2 norm instead [3]. Another
person in stackoverflow seems to suggest the use of per-vertex tolerance
comparison, which is essentially the L∞ norm [4]. The L1 norm ||E||1 between
two (rank) vectors r and s is calculated as Σ|rn - sn|, or as the sum of absolute
errors. The L2 norm ||E||2 is calculated as √Σ|rn - sn|2
, or as the square-root of
the sum of squared errors (euclidean distance between the two vectors).
The L∞ norm ||E||∞ is calculated as max(|rn - sn|), or as the maximum of
absolute errors.
This experiment was for comparing the performance between PageRank
computation with L1, L2 and L∞ norms as convergence check, for damping
factor α = 0.85, and tolerance τ = 10-6
. The input graphs, system used, and
the rest of the experimental process is similar to that of the first experiment.
Additionally, the execution time of each test case is measured using
std::chrono::high_performance_timer. This is done 5 times for each test case,
and timings are averaged (AM).
From the results, as seen in figures 3.1 and 3.2, it is clear that PageRank
computation with L∞ norm as convergence check is the fastest, quickly
followed by L2 norm, and finally L1 norm. Thus, when comparing two or
more approaches for an iterative algorithm, it is important to ensure that all
of them use the same error function as convergence check (and the same
parameter values). This would help ensure a level ground for a good relative
performance comparison.
Also note in figures 3.1 and 3.2 that PageRank computation with L∞ norm as
convergence check completes in a single iteration for all the road networks
(ending with _osm). This is likely because it is calculated as ||E||∞ = max(|rn -
6
sn|), and depending upon the order (number of vertices) N of the graph (those
graphs are quite large), the maximum rank change for any single vertex does
not exceed the tolerance τ value of 10-6
.
Figure 3.1: Iterations needed for PageRank computation for various graphs with
damping factor α = 0.85, and tolerance τ = 10-6
, with L1, L2, and L∞ norms as
convergence check.
Figure 3.2: Relative iterations for PageRank computations for various graphs with L1,
L2, and L∞ norms as convergence check, with respect to L1 norm as baseline
(damping factor α = 0.85, and tolerance τ = 10-6
).
7
AM-RATIO 4,645 ms 2,052 ms 1,070 ms 1.00 0.44 0.23 4.34 1.92 1.00
GM-RATIO 1,291 ms 626 ms 236 ms 1.00 0.48 0.18 5.46 2.65 1.00
HM-RATIO 351 ms 198 ms 118 ms 1.00 0.56 0.34 2.98 1.68 1.00
RATIO-AM 1.00 0.50 0.33 15.92 5.66 1.00
RATIO-GM 1.00 0.48 0.18 5.46 2.65 1.00
RATIO-HM 1.00 0.47 0.06 3.03 1.75 1.00
Table 3.1: Mean times for PageRank computation with L1, L2, and L∞ norms as
convergence check (first 3 columns), ratios relative to L1 norm (second 3 columns),
and then to L∞ norm (last 3 columns). Blue rows are unaffected by baseline choice,
purple cells are close to ratios for mean iterations (see table 3.2), and red cells are
significantly different.
AM-RATIO 57.29 28.82 18.94 1.00 0.50 0.33 3.02 1.52 1.00
GM-RATIO 56.91 27.50 10.21 1.00 0.48 0.18 5.57 2.69 1.00
HM-RATIO 56.51 26.32 3.55 1.00 0.47 0.06 15.91 7.41 1.00
RATIO-AM 1.00 0.50 0.33 17.61 6.08 1.00
RATIO-GM 1.00 0.48 0.18 5.57 2.69 1.00
RATIO-HM 1.00 0.46 0.06 3.00 1.74 1.00
Table 3.2: Mean iterations for PageRank computation with L1, L2, and L∞ norms as
convergence check (first 3 columns), ratios relative to L1 norm (second 3 columns),
and then to L∞ norm (last 3 columns). Blue rows are unaffected by baseline choice,
purple cells are close to ratios for mean iterations (see table 3.1), and red cells are
significantly different.
In order to obtain a composite relative performance ratio, the six different
methods mentioned above are calculated, for both PageRank computation
time and iterations. They are listed in table 3.1 and 3.2 respectively. The
ratios are calculated with both L1, and L∞ norms as baseline. Methods which
calculate ratios at the end are not affected by the choice of baseline, as
expected. However, not only is RATIO-GM unaffected as well, it is also
identical to GM-RATIO, due to its mathematical property. It is observed that
RATIO-AM and RATIO-HM are affected by the choice of baseline, while
RATIO-GM is unaffected. RATIO-HM, RATIO-GM, and RATIO-AM with L1
norm as baseline are similar for both time and iterations, but HM-RATIO is
significantly different. This indicates GM-RATIO to be the most stable
8
composite relative performance ratio, followed by AM-RATIO. In fact,
weighted GM-RATIO is used by SPECviewperf® [2], as mentioned above.
Semantically, GM-RATIO comparison gives equal importance to the relative
performance of each test case (graph), while an AM-RATIO comparison gives
equal importance to magnitude (time/iterations) of all test cases (or simply, it
gives higher importance to test cases with larger graphs).
4. Adjusting Tolerance
Similar to the damping factor α and the error function used for convergence
check, adjusting the value of tolerance τ can have a significant effect. This
experiment was for comparing the performance between PageRank
computation with L1, L2 and L∞ norms as convergence check, for various
tolerance τ values ranging from 10-0
to 10-10
(10-0
, 5×10-0
, 10-1
, 5×10-1
, ...).
The input graphs, system used, and the rest of the experimental process is
similar to that of the first experiment.
For various graphs, some of which are shown in figures 4.1 and 4.2, it is
observed that PageRank computation with L1, L2, or L∞ norm as
convergence check suffers from sensitivity issues beyond certain (smaller)
tolerance τ values, causing the computation to halt at maximum iteration
limit (500) without convergence. As tolerance τ is decreased from 10-0
to
10-10
, L1 norm is the first to suffer from this issue, followed by L2 and L∞
norms (except road networks). This sensitivity issue was recognized by the
fact that a given approach abruptly takes 500 iterations for the next lower
tolerance τ value.
It is also observed, as shown in figure 4.2, that PageRank computation with
L∞ norm as convergence check completes in just one iteration (even for
tolerance τ ≥ 10-6
) for large graphs (road networks). This again, as mentioned
above, is likely because the maximum rank change for any single vertex for
L∞ norm, and the sum of squares of total rank change for all vertices, is quite
low for such large graphs. Thus, it does not exceed the given tolerance τ
value, causing a single iteration convergence.
9
Figure 4.1: Iterations taken for PageRank computation of the web-BerkStan graph,
with L1, L2, and L∞ norms used as convergence check. Up till tolerance τ = 5×10-2
,
L2 and L∞ norms converge in just a single iteration. From tolerance τ = 10-8
, L1 norm
begins to suffer from sensitivity issues, followed by L2 and L∞ norms at 10-9
.
Figure 4.2: Iterations taken for PageRank computation of the asia_osm graph, with
L1, L2, and L∞ norms used as convergence check. Until tolerance τ = 10-7
, the L∞
norm converges in just one iteration.
On average, PageRank computation with L∞ norm as the error function is the
fastest, quickly followed by L2 norm, and then L1 norm. This is the case
with both geometric mean (GM) and arithmetic mean (AM) comparisons of
10
iterations needed for convergence with each of the three error functions, as
shown in figures 4.3 and 4.4. In fact, this trend is observed with each of the
individual graphs separately, although not shown here.
Figure 4.3: Geometric mean iterations taken for PageRank computation with L1, L2
and L∞ norms as convergence check, and tolerance τ adjusted from 10-0
to 10-10
.
Figure 4.4: Arithmetic mean iterations taken for PageRank computation with L1, L2
and L∞ norms as convergence check, and tolerance τ adjusted from 10-0
to 10-10
.
11
Figure 4.5: Relative GM iterations taken for PageRank computation with L1, L2 and
L∞ norms as convergence check, and tolerance τ adjusted from 10-0
to 10-10
.
Figure 4.6: Relative GM iterations taken for PageRank computation with L1, L2 and
L∞ norms as convergence check, and tolerance τ adjusted from 10-0
to 10-10
.
Based on GM-RATIO comparison, the relative iterations between PageRank
computation with L1, L2, and L∞ norm as convergence check is 1.00 : 0.30 :
0.20. Hence L2 norm is on average 70% faster than L1 norm, and L∞ norm is
33% faster than L2 norm. This ratio is calculated by first finding the GM of
iterations based on each error function for each tolerance τ value separately.
12
These tolerance τ specific means are then combined with GM to obtain a
single mean value for each error function (norm). The GM-RATIO is then the
ratio of each norm with respect to the L∞ norm. The variation of tolerance τ
specific means with L∞ norm as baseline for various tolerance τ values is
shown in figure 4.5.
On the other hand, based on AM-RATIO comparison, the relative iterations
between PageRank computation with L1, L2, and L∞ norm as convergence
check is 1.00 : 0.39 : 0.31. Hence, L2 norm is on average 61% faster than L1
norm, and L∞ norm is 26% faster than L2 norm. This ratio is calculated in a
manner similar to that of GM-RATIO, except that it uses AM instead of GM.
The variation of tolerance τ specific means with L∞ norm as baseline for
various tolerance τ values is shown in figure 4.6.
5. Conclusion
Parameter values can have a significant effect on performance, as seen in
these experiments. Different error functions converge at different rates, and
which of them converges faster depends upon the tolerance τ value.
Iteration count needs to be checked in order to ensure that no approach is
suffering from sensitivity issues, or is leading to a single iteration
convergence. Finally, the relative performance comparison method affects
which results get more importance, and which do not, in the final average.
Taking note of each of these points, when comparing iterative algorithms,
will thus ensure that the performance results are accurate and useful. The
links to source code, along with data sheets and charts, for adjusting
damping factor [5], error function [6], and tolerance [7] are included in
references.
References
[1] A. Langville and C. Meyer, “Deeper Inside PageRank,” Internet Math.,
vol. 1, no. 3, pp. 335–380, Jan. 2004, doi:
10.1080/15427951.2004.10129091.
[2] B. Licea-Kane, “SPECviewperf® 7.1 -- Weighted Geometric Mean.”
13
https://www.spec.org/gwpg/gpc.static/geometric.html (accessed Aug.
05, 2021).
[3] A. Fender, “nvgraph/pagerank.cu at main · rapidsai/nvgraph.”
https://github.com/rapidsai/nvgraph/blob/main/cpp/src/pagerank.cu#L149
(accessed Aug. 06, 2021).
[4] N. Upreti, “How to check for Page Rank convergence?”
https://stackoverflow.com/a/29321153/1413259 (accessed Aug. 06,
2021).
[5] S. Sahu, “puzzlef/pagerank-adjust-damping-factor: Comparing the effect
of using different values of damping factor, with PageRank (pull, CSR).”
https://github.com/puzzlef/pagerank-adjust-damping-factor (accessed
Aug. 06, 2021).
[6] S. Sahu, “puzzlef/pagerank-adjust-tolerance-function: Comparing the
effect of using different functions for convergence check, with PageRank
(pull, CSR).”
https://github.com/puzzlef/pagerank-adjust-tolerance-function (accessed
Aug. 06, 2021).
[7] S. Sahu, “puzzlef/pagerank-adjust-tolerance: Comparing the effect of
using different values of tolerance, with PageRank (pull, CSR).”
https://github.com/puzzlef/pagerank-adjust-tolerance (accessed Aug. 06,
2021).

More Related Content

What's hot

X bar and-r_charts
X bar and-r_chartsX bar and-r_charts
X bar and-r_charts
Sadvachan Mishra
 
Fibonacci_Hubble
Fibonacci_HubbleFibonacci_Hubble
Fibonacci_Hubble
Marc King
 
Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]
Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]
Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]
AI Robotics KR
 
Effect of stepwise adjustment of Damping factor upon PageRank : REPORT
Effect of stepwise adjustment of Damping factor upon PageRank : REPORTEffect of stepwise adjustment of Damping factor upon PageRank : REPORT
Effect of stepwise adjustment of Damping factor upon PageRank : REPORT
Subhajit Sahu
 
Chapter 3 mathematical modeling of dynamic system
Chapter 3 mathematical modeling of dynamic systemChapter 3 mathematical modeling of dynamic system
Chapter 3 mathematical modeling of dynamic system
LenchoDuguma
 
Receptor kinetics Laplace transform method Word
Receptor kinetics Laplace transform method WordReceptor kinetics Laplace transform method Word
Receptor kinetics Laplace transform method Word
Sam Hoare
 
AR model
AR modelAR model
AR model
Naveen Kumar
 
Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...
Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...
Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...
AI Robotics KR
 
A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...
A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...
A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...
IJERA Editor
 
Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]
Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]
Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]
AI Robotics KR
 
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
AI Robotics KR
 
case study of curve fitting
case study of curve fittingcase study of curve fitting
case study of curve fitting
Adarsh Patel
 
Estimation Theory Class (Summary and Revision)
Estimation Theory Class (Summary and Revision)Estimation Theory Class (Summary and Revision)
Estimation Theory Class (Summary and Revision)
Ahmad Gomaa
 
REYLEIGH’S METHOD,BUCKINGHAM π-THEOREM
REYLEIGH’S METHOD,BUCKINGHAM  π-THEOREMREYLEIGH’S METHOD,BUCKINGHAM  π-THEOREM
REYLEIGH’S METHOD,BUCKINGHAM π-THEOREM
Amiraj College Of Engineering And Technology
 
Training the neural network using levenberg marquardt’s algorithm to optimize
Training the neural network using levenberg marquardt’s algorithm to optimizeTraining the neural network using levenberg marquardt’s algorithm to optimize
Training the neural network using levenberg marquardt’s algorithm to optimize
IAEME Publication
 
Isen 614 project presentation
Isen 614 project presentationIsen 614 project presentation
Isen 614 project presentation
Vanshaj Handoo
 
Statistical computing with r estatistica - maria l. rizzo
Statistical computing with r   estatistica - maria l. rizzoStatistical computing with r   estatistica - maria l. rizzo
Statistical computing with r estatistica - maria l. rizzo
André Oliveira Souza
 
Modern Control - Lec 02 - Mathematical Modeling of Systems
Modern Control - Lec 02 - Mathematical Modeling of SystemsModern Control - Lec 02 - Mathematical Modeling of Systems
Modern Control - Lec 02 - Mathematical Modeling of Systems
Amr E. Mohamed
 
A Comparative study of locality Preserving Projection & Principle Component A...
A Comparative study of locality Preserving Projection & Principle Component A...A Comparative study of locality Preserving Projection & Principle Component A...
A Comparative study of locality Preserving Projection & Principle Component A...
RAHUL WAGAJ
 
Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]
Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]
Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]
AI Robotics KR
 

What's hot (20)

X bar and-r_charts
X bar and-r_chartsX bar and-r_charts
X bar and-r_charts
 
Fibonacci_Hubble
Fibonacci_HubbleFibonacci_Hubble
Fibonacci_Hubble
 
Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]
Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]
Sensor Fusion Study - Ch7. Kalman Filter Generalizations [김영범]
 
Effect of stepwise adjustment of Damping factor upon PageRank : REPORT
Effect of stepwise adjustment of Damping factor upon PageRank : REPORTEffect of stepwise adjustment of Damping factor upon PageRank : REPORT
Effect of stepwise adjustment of Damping factor upon PageRank : REPORT
 
Chapter 3 mathematical modeling of dynamic system
Chapter 3 mathematical modeling of dynamic systemChapter 3 mathematical modeling of dynamic system
Chapter 3 mathematical modeling of dynamic system
 
Receptor kinetics Laplace transform method Word
Receptor kinetics Laplace transform method WordReceptor kinetics Laplace transform method Word
Receptor kinetics Laplace transform method Word
 
AR model
AR modelAR model
AR model
 
Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...
Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...
Sensor Fusion Study - Ch10. Additional topics in kalman filter [Stella Seoyeo...
 
A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...
A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...
A Comparative Study of Acoustic Echo Cancellation Algorithms in Sparse Impuls...
 
Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]
Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]
Sensor Fusion Study - Ch13. Nonlinear Kalman Filtering [Ahn Min Sung]
 
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
 
case study of curve fitting
case study of curve fittingcase study of curve fitting
case study of curve fitting
 
Estimation Theory Class (Summary and Revision)
Estimation Theory Class (Summary and Revision)Estimation Theory Class (Summary and Revision)
Estimation Theory Class (Summary and Revision)
 
REYLEIGH’S METHOD,BUCKINGHAM π-THEOREM
REYLEIGH’S METHOD,BUCKINGHAM  π-THEOREMREYLEIGH’S METHOD,BUCKINGHAM  π-THEOREM
REYLEIGH’S METHOD,BUCKINGHAM π-THEOREM
 
Training the neural network using levenberg marquardt’s algorithm to optimize
Training the neural network using levenberg marquardt’s algorithm to optimizeTraining the neural network using levenberg marquardt’s algorithm to optimize
Training the neural network using levenberg marquardt’s algorithm to optimize
 
Isen 614 project presentation
Isen 614 project presentationIsen 614 project presentation
Isen 614 project presentation
 
Statistical computing with r estatistica - maria l. rizzo
Statistical computing with r   estatistica - maria l. rizzoStatistical computing with r   estatistica - maria l. rizzo
Statistical computing with r estatistica - maria l. rizzo
 
Modern Control - Lec 02 - Mathematical Modeling of Systems
Modern Control - Lec 02 - Mathematical Modeling of SystemsModern Control - Lec 02 - Mathematical Modeling of Systems
Modern Control - Lec 02 - Mathematical Modeling of Systems
 
A Comparative study of locality Preserving Projection & Principle Component A...
A Comparative study of locality Preserving Projection & Principle Component A...A Comparative study of locality Preserving Projection & Principle Component A...
A Comparative study of locality Preserving Projection & Principle Component A...
 
Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]
Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]
Sensor Fusion Study - Ch8. The Continuous-Time Kalman Filter [이해구]
 

Similar to Adjusting PageRank parameters and comparing results : REPORT

Ijrdt11 140004
Ijrdt11 140004Ijrdt11 140004
Ijrdt11 140004
Ijrdt Journal
 
A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...
A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...
A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...
IRJET Journal
 
Simple lin regress_inference
Simple lin regress_inferenceSimple lin regress_inference
Simple lin regress_inference
Kemal İnciroğlu
 
Building the Professional of 2020: An Approach to Business Change Process Int...
Building the Professional of 2020: An Approach to Business Change Process Int...Building the Professional of 2020: An Approach to Business Change Process Int...
Building the Professional of 2020: An Approach to Business Change Process Int...
Dr Harris Apostolopoulos EMBA, PfMP, PgMP, PMP, IPMO-E
 
A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...
A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...
A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...
Waqas Tariq
 
R analysis of covariance
R   analysis of covarianceR   analysis of covariance
R analysis of covariance
Learnbay Datascience
 
2014-mo444-practical-assignment-04-paulo_faria
2014-mo444-practical-assignment-04-paulo_faria2014-mo444-practical-assignment-04-paulo_faria
2014-mo444-practical-assignment-04-paulo_faria
Paulo Faria
 
RS
RSRS
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTESAdjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
Subhajit Sahu
 
Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...
Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...
Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...
Subhajit Sahu
 
A Novel Cosine Approximation for High-Speed Evaluation of DCT
A Novel Cosine Approximation for High-Speed Evaluation of DCTA Novel Cosine Approximation for High-Speed Evaluation of DCT
A Novel Cosine Approximation for High-Speed Evaluation of DCT
CSCJournals
 
working with python
working with pythonworking with python
working with python
bhavesh lande
 
Integral method to analyze reaction kinetics
Integral method to analyze reaction kineticsIntegral method to analyze reaction kinetics
Integral method to analyze reaction kinetics
varshabhi27
 
Design and Implementation of Model Reference Adaptive Controller using Coeffi...
Design and Implementation of Model Reference Adaptive Controller using Coeffi...Design and Implementation of Model Reference Adaptive Controller using Coeffi...
Design and Implementation of Model Reference Adaptive Controller using Coeffi...
IOSR Journals
 
An econometric model for Linear Regression using Statistics
An econometric model for Linear Regression using StatisticsAn econometric model for Linear Regression using Statistics
An econometric model for Linear Regression using Statistics
IRJET Journal
 
A study of the Behavior of Floating-Point Errors
A study of the Behavior of Floating-Point ErrorsA study of the Behavior of Floating-Point Errors
A study of the Behavior of Floating-Point Errors
ijpla
 
2018867974 sulaim (2)
2018867974 sulaim (2)2018867974 sulaim (2)
2018867974 sulaim (2)
sulaim_qais
 
Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...
Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...
Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...
University Utara Malaysia
 
Doe introductionh
Doe introductionhDoe introductionh
Doe introductionh
sumandaspersonal
 
(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...
(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...
(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...
Naoki Shibata
 

Similar to Adjusting PageRank parameters and comparing results : REPORT (20)

Ijrdt11 140004
Ijrdt11 140004Ijrdt11 140004
Ijrdt11 140004
 
A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...
A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...
A Study of Training and Blind Equalization Algorithms for Quadrature Amplitud...
 
Simple lin regress_inference
Simple lin regress_inferenceSimple lin regress_inference
Simple lin regress_inference
 
Building the Professional of 2020: An Approach to Business Change Process Int...
Building the Professional of 2020: An Approach to Business Change Process Int...Building the Professional of 2020: An Approach to Business Change Process Int...
Building the Professional of 2020: An Approach to Business Change Process Int...
 
A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...
A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...
A Computationally Efficient Algorithm to Solve Generalized Method of Moments ...
 
R analysis of covariance
R   analysis of covarianceR   analysis of covariance
R analysis of covariance
 
2014-mo444-practical-assignment-04-paulo_faria
2014-mo444-practical-assignment-04-paulo_faria2014-mo444-practical-assignment-04-paulo_faria
2014-mo444-practical-assignment-04-paulo_faria
 
RS
RSRS
RS
 
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTESAdjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
 
Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...
Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...
Algorithmic optimizations for Dynamic Monolithic PageRank (from STICD) : SHOR...
 
A Novel Cosine Approximation for High-Speed Evaluation of DCT
A Novel Cosine Approximation for High-Speed Evaluation of DCTA Novel Cosine Approximation for High-Speed Evaluation of DCT
A Novel Cosine Approximation for High-Speed Evaluation of DCT
 
working with python
working with pythonworking with python
working with python
 
Integral method to analyze reaction kinetics
Integral method to analyze reaction kineticsIntegral method to analyze reaction kinetics
Integral method to analyze reaction kinetics
 
Design and Implementation of Model Reference Adaptive Controller using Coeffi...
Design and Implementation of Model Reference Adaptive Controller using Coeffi...Design and Implementation of Model Reference Adaptive Controller using Coeffi...
Design and Implementation of Model Reference Adaptive Controller using Coeffi...
 
An econometric model for Linear Regression using Statistics
An econometric model for Linear Regression using StatisticsAn econometric model for Linear Regression using Statistics
An econometric model for Linear Regression using Statistics
 
A study of the Behavior of Floating-Point Errors
A study of the Behavior of Floating-Point ErrorsA study of the Behavior of Floating-Point Errors
A study of the Behavior of Floating-Point Errors
 
2018867974 sulaim (2)
2018867974 sulaim (2)2018867974 sulaim (2)
2018867974 sulaim (2)
 
Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...
Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...
Strategic Oscillation for Exploitation and Exploration of ACS Algorithm for J...
 
Doe introductionh
Doe introductionhDoe introductionh
Doe introductionh
 
(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...
(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...
(Slides) Efficient Evaluation Methods of Elementary Functions Suitable for SI...
 

More from Subhajit Sahu

About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...
About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...
About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...
Subhajit Sahu
 
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Subhajit Sahu
 
Adjusting Bitset for graph : SHORT REPORT / NOTES
Adjusting Bitset for graph : SHORT REPORT / NOTESAdjusting Bitset for graph : SHORT REPORT / NOTES
Adjusting Bitset for graph : SHORT REPORT / NOTES
Subhajit Sahu
 
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Subhajit Sahu
 
Adjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTESAdjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTES
Subhajit Sahu
 
Experiments with Primitive operations : SHORT REPORT / NOTES
Experiments with Primitive operations : SHORT REPORT / NOTESExperiments with Primitive operations : SHORT REPORT / NOTES
Experiments with Primitive operations : SHORT REPORT / NOTES
Subhajit Sahu
 
PageRank Experiments : SHORT REPORT / NOTES
PageRank Experiments : SHORT REPORT / NOTESPageRank Experiments : SHORT REPORT / NOTES
PageRank Experiments : SHORT REPORT / NOTES
Subhajit Sahu
 
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
Subhajit Sahu
 
DyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTES
DyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTESDyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTES
DyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTES
Subhajit Sahu
 
Shared memory Parallelism (NOTES)
Shared memory Parallelism (NOTES)Shared memory Parallelism (NOTES)
Shared memory Parallelism (NOTES)
Subhajit Sahu
 
A Dynamic Algorithm for Local Community Detection in Graphs : NOTES
A Dynamic Algorithm for Local Community Detection in Graphs : NOTESA Dynamic Algorithm for Local Community Detection in Graphs : NOTES
A Dynamic Algorithm for Local Community Detection in Graphs : NOTES
Subhajit Sahu
 
Scalable Static and Dynamic Community Detection Using Grappolo : NOTES
Scalable Static and Dynamic Community Detection Using Grappolo : NOTESScalable Static and Dynamic Community Detection Using Grappolo : NOTES
Scalable Static and Dynamic Community Detection Using Grappolo : NOTES
Subhajit Sahu
 
Application Areas of Community Detection: A Review : NOTES
Application Areas of Community Detection: A Review : NOTESApplication Areas of Community Detection: A Review : NOTES
Application Areas of Community Detection: A Review : NOTES
Subhajit Sahu
 
Community Detection on the GPU : NOTES
Community Detection on the GPU : NOTESCommunity Detection on the GPU : NOTES
Community Detection on the GPU : NOTES
Subhajit Sahu
 
Survey for extra-child-process package : NOTES
Survey for extra-child-process package : NOTESSurvey for extra-child-process package : NOTES
Survey for extra-child-process package : NOTES
Subhajit Sahu
 
Dynamic Batch Parallel Algorithms for Updating PageRank : POSTER
Dynamic Batch Parallel Algorithms for Updating PageRank : POSTERDynamic Batch Parallel Algorithms for Updating PageRank : POSTER
Dynamic Batch Parallel Algorithms for Updating PageRank : POSTER
Subhajit Sahu
 
Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...
Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...
Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...
Subhajit Sahu
 
Fast Incremental Community Detection on Dynamic Graphs : NOTES
Fast Incremental Community Detection on Dynamic Graphs : NOTESFast Incremental Community Detection on Dynamic Graphs : NOTES
Fast Incremental Community Detection on Dynamic Graphs : NOTES
Subhajit Sahu
 
Can you fix farming by going back 8000 years : NOTES
Can you fix farming by going back 8000 years : NOTESCan you fix farming by going back 8000 years : NOTES
Can you fix farming by going back 8000 years : NOTES
Subhajit Sahu
 
HITS algorithm : NOTES
HITS algorithm : NOTESHITS algorithm : NOTES
HITS algorithm : NOTES
Subhajit Sahu
 

More from Subhajit Sahu (20)

About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...
About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...
About TrueTime, Spanner, Clock synchronization, CAP theorem, Two-phase lockin...
 
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
 
Adjusting Bitset for graph : SHORT REPORT / NOTES
Adjusting Bitset for graph : SHORT REPORT / NOTESAdjusting Bitset for graph : SHORT REPORT / NOTES
Adjusting Bitset for graph : SHORT REPORT / NOTES
 
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
 
Adjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTESAdjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTES
 
Experiments with Primitive operations : SHORT REPORT / NOTES
Experiments with Primitive operations : SHORT REPORT / NOTESExperiments with Primitive operations : SHORT REPORT / NOTES
Experiments with Primitive operations : SHORT REPORT / NOTES
 
PageRank Experiments : SHORT REPORT / NOTES
PageRank Experiments : SHORT REPORT / NOTESPageRank Experiments : SHORT REPORT / NOTES
PageRank Experiments : SHORT REPORT / NOTES
 
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
 
DyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTES
DyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTESDyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTES
DyGraph: A Dynamic Graph Generator and Benchmark Suite : NOTES
 
Shared memory Parallelism (NOTES)
Shared memory Parallelism (NOTES)Shared memory Parallelism (NOTES)
Shared memory Parallelism (NOTES)
 
A Dynamic Algorithm for Local Community Detection in Graphs : NOTES
A Dynamic Algorithm for Local Community Detection in Graphs : NOTESA Dynamic Algorithm for Local Community Detection in Graphs : NOTES
A Dynamic Algorithm for Local Community Detection in Graphs : NOTES
 
Scalable Static and Dynamic Community Detection Using Grappolo : NOTES
Scalable Static and Dynamic Community Detection Using Grappolo : NOTESScalable Static and Dynamic Community Detection Using Grappolo : NOTES
Scalable Static and Dynamic Community Detection Using Grappolo : NOTES
 
Application Areas of Community Detection: A Review : NOTES
Application Areas of Community Detection: A Review : NOTESApplication Areas of Community Detection: A Review : NOTES
Application Areas of Community Detection: A Review : NOTES
 
Community Detection on the GPU : NOTES
Community Detection on the GPU : NOTESCommunity Detection on the GPU : NOTES
Community Detection on the GPU : NOTES
 
Survey for extra-child-process package : NOTES
Survey for extra-child-process package : NOTESSurvey for extra-child-process package : NOTES
Survey for extra-child-process package : NOTES
 
Dynamic Batch Parallel Algorithms for Updating PageRank : POSTER
Dynamic Batch Parallel Algorithms for Updating PageRank : POSTERDynamic Batch Parallel Algorithms for Updating PageRank : POSTER
Dynamic Batch Parallel Algorithms for Updating PageRank : POSTER
 
Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...
Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...
Abstract for IPDPS 2022 PhD Forum on Dynamic Batch Parallel Algorithms for Up...
 
Fast Incremental Community Detection on Dynamic Graphs : NOTES
Fast Incremental Community Detection on Dynamic Graphs : NOTESFast Incremental Community Detection on Dynamic Graphs : NOTES
Fast Incremental Community Detection on Dynamic Graphs : NOTES
 
Can you fix farming by going back 8000 years : NOTES
Can you fix farming by going back 8000 years : NOTESCan you fix farming by going back 8000 years : NOTES
Can you fix farming by going back 8000 years : NOTES
 
HITS algorithm : NOTES
HITS algorithm : NOTESHITS algorithm : NOTES
HITS algorithm : NOTES
 

Recently uploaded

Google ML-Kit - Understanding on-device machine learning
Google ML-Kit - Understanding on-device machine learningGoogle ML-Kit - Understanding on-device machine learning
Google ML-Kit - Understanding on-device machine learning
VishrutGoyani1
 
Comprehensive Vulnerability Assessments Process _ Aardwolf Security.docx
Comprehensive Vulnerability Assessments Process _ Aardwolf Security.docxComprehensive Vulnerability Assessments Process _ Aardwolf Security.docx
Comprehensive Vulnerability Assessments Process _ Aardwolf Security.docx
Aardwolf Security
 
UMiami degree offer diploma Transcript
UMiami degree offer diploma TranscriptUMiami degree offer diploma Transcript
UMiami degree offer diploma Transcript
attueb
 
Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...
Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...
Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...
ashiklo9823
 
bangalore Girls call 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
bangalore Girls call  👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Deliverybangalore Girls call  👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
bangalore Girls call 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
sunilverma7884
 
HIRE A HACKER FOR CHEATING HUSBAND/WIFE)
HIRE A HACKER FOR CHEATING HUSBAND/WIFE)HIRE A HACKER FOR CHEATING HUSBAND/WIFE)
HIRE A HACKER FOR CHEATING HUSBAND/WIFE)
josephinedrea942
 
ThaiPy meetup - Indexes and Django
ThaiPy meetup - Indexes and DjangoThaiPy meetup - Indexes and Django
ThaiPy meetup - Indexes and Django
akshesh doshi
 
VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...
VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...
VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...
jealousviolet
 
Software development... for all? (keynote at ICSOFT'2024)
Software development... for all? (keynote at ICSOFT'2024)Software development... for all? (keynote at ICSOFT'2024)
Software development... for all? (keynote at ICSOFT'2024)
miso_uam
 
NYGGS 360: A Complete ERP for Construction Innovation
NYGGS 360: A Complete ERP for Construction InnovationNYGGS 360: A Complete ERP for Construction Innovation
NYGGS 360: A Complete ERP for Construction Innovation
NYGGS Construction ERP Software
 
Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...
Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...
Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...
902basic
 
ERP Software Solutions Provider in Coimbatore
ERP Software Solutions Provider in CoimbatoreERP Software Solutions Provider in Coimbatore
ERP Software Solutions Provider in Coimbatore
Nextskill Technologies
 
Odoo E-commerce website development guides
Odoo E-commerce website development guidesOdoo E-commerce website development guides
Odoo E-commerce website development guides
jhkdigitalmarketing
 
Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...
Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...
Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...
dream girl
 
Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...
Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...
Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...
revolutionary575
 
AI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdf
AI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdfAI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdf
AI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdf
Daniel Zivkovic
 
Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...
Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...
Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...
aslasdfmkhan4750
 
Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...
Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...
Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...
shanihomely
 
Il Data Streaming per un’AI real-time di nuova generazione
Il Data Streaming per un’AI real-time di nuova generazioneIl Data Streaming per un’AI real-time di nuova generazione
Il Data Streaming per un’AI real-time di nuova generazione
confluent
 
Top Chinese Government-backed APT Groups
Top Chinese Government-backed APT GroupsTop Chinese Government-backed APT Groups
Top Chinese Government-backed APT Groups
SOCRadar
 

Recently uploaded (20)

Google ML-Kit - Understanding on-device machine learning
Google ML-Kit - Understanding on-device machine learningGoogle ML-Kit - Understanding on-device machine learning
Google ML-Kit - Understanding on-device machine learning
 
Comprehensive Vulnerability Assessments Process _ Aardwolf Security.docx
Comprehensive Vulnerability Assessments Process _ Aardwolf Security.docxComprehensive Vulnerability Assessments Process _ Aardwolf Security.docx
Comprehensive Vulnerability Assessments Process _ Aardwolf Security.docx
 
UMiami degree offer diploma Transcript
UMiami degree offer diploma TranscriptUMiami degree offer diploma Transcript
UMiami degree offer diploma Transcript
 
Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...
Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...
Vip Girls Call ServiCe Hyderabad 0000000000 Pooja Best High Class Hyderabad A...
 
bangalore Girls call 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
bangalore Girls call  👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Deliverybangalore Girls call  👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
bangalore Girls call 👀 XXXXXXXXXXX 👀 Rs.9.5 K Cash Payment With Room Delivery
 
HIRE A HACKER FOR CHEATING HUSBAND/WIFE)
HIRE A HACKER FOR CHEATING HUSBAND/WIFE)HIRE A HACKER FOR CHEATING HUSBAND/WIFE)
HIRE A HACKER FOR CHEATING HUSBAND/WIFE)
 
ThaiPy meetup - Indexes and Django
ThaiPy meetup - Indexes and DjangoThaiPy meetup - Indexes and Django
ThaiPy meetup - Indexes and Django
 
VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...
VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...
VVIP Girls Call Mumbai 9910780858 Provide Best And Top Girl Service And No1 i...
 
Software development... for all? (keynote at ICSOFT'2024)
Software development... for all? (keynote at ICSOFT'2024)Software development... for all? (keynote at ICSOFT'2024)
Software development... for all? (keynote at ICSOFT'2024)
 
NYGGS 360: A Complete ERP for Construction Innovation
NYGGS 360: A Complete ERP for Construction InnovationNYGGS 360: A Complete ERP for Construction Innovation
NYGGS 360: A Complete ERP for Construction Innovation
 
Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...
Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...
Private Girls Call Navi Mumbai 🛵🚡9820252231 💃 Choose Best And Top Girl Servic...
 
ERP Software Solutions Provider in Coimbatore
ERP Software Solutions Provider in CoimbatoreERP Software Solutions Provider in Coimbatore
ERP Software Solutions Provider in Coimbatore
 
Odoo E-commerce website development guides
Odoo E-commerce website development guidesOdoo E-commerce website development guides
Odoo E-commerce website development guides
 
Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...
Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...
Russian Girls Call Mumbai 🛵🚡9833363713 💃 Choose Best And Top Girl Service And...
 
Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...
Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...
Busty Girls Call Mumbai 9930245274 Unlimited Short Providing Girls Service Av...
 
AI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdf
AI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdfAI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdf
AI - Your Startup Sidekick (Leveraging AI to Bootstrap a Lean Startup).pdf
 
Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...
Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...
Independent Girls Call ServiCe Hyderabad 0000000000 Tanisha Best High Class H...
 
Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...
Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...
Russian Girls Call Mumbai 🎈🔥9930687706 🔥💋🎈 Provide Best And Top Girl Service ...
 
Il Data Streaming per un’AI real-time di nuova generazione
Il Data Streaming per un’AI real-time di nuova generazioneIl Data Streaming per un’AI real-time di nuova generazione
Il Data Streaming per un’AI real-time di nuova generazione
 
Top Chinese Government-backed APT Groups
Top Chinese Government-backed APT GroupsTop Chinese Government-backed APT Groups
Top Chinese Government-backed APT Groups
 

Adjusting PageRank parameters and comparing results : REPORT

  • 1. 1 Adjusting PageRank parameters and Comparing results Subhajit Sahu1 , Kishore Kothapalli1 , Dip Sankar Banerjee2 1 International Institute of Information Technology, Hyderabad 2 Indian Institute of Technology, Jodhpur Abstract — The effect of adjusting damping factor α and tolerance τ on iterations needed for PageRank computation is studied here. Relative performance of PageRank computation with L1, L2, and L∞ norms used as convergence check, are also compared with six possible mean ratios. It is observed that increasing the damping factor α linearly increases the iterations needed almost exponentially. On the other hand, decreasing the tolerance τ exponentially decreases the iterations needed almost exponentially. On average, PageRank with L∞ norm as convergence check is the fastest, quickly followed by L2 norm, and then L1 norm. For large graphs, above certain tolerance τ values, convergence can occur in a single iteration. On the contrary, below certain tolerance τ values, sensitivity issues can begin to appear, causing computation to halt at maximum iteration limit without convergence. The six mean ratios for relative performance comparison are based on arithmetic, geometric, and harmonic mean, as well as the order of ratio calculation. Among them GM-RATIO, geometric mean followed by ratio calculation, is found to be most stable, followed by AM-RATIO. Index terms — PageRank algorithm, Parameter adjustment, Convergence function, Sensitivity issues, Relative performance comparison. 1. Introduction Web graphs unaltered are reducible, and thus the rate of convergence of the power-iteration method is the rate at which αk → 0, where α is the damping factor, and k is the iteration count. An estimate of the number of iterations needed to converge to a tolerance τ is log10 τ / log10 α [1]. For τ = 10-6 and α = 0.85, it can take roughly 85 iterations to converge. For α = 0.95, and α = 0.75, with the same tolerance τ = 10-6 , it takes roughly 269 and 48 iterations respectively. For τ = 10-9 , and τ = 10-3 , with the same damping factor α =
  • 2. 2 0.85, it takes roughly 128 and 43 iterations respectively. Thus, adjusting the damping factor or the tolerance parameters of the PageRank algorithm can have a significant effect on the convergence rate. However, once results for various test cases are obtained, there exist multiple methods to obtain a composite relative performance ratio. Consider, for example, three approaches a, b, and c, with 3 test runs for each of the three approaches, labeled a1, a2, a3, b1, b2, b3, c1, c2, c3. One method to get a composite ratio between the three approaches would be to find the relative performance ratio of each approach with respect to a baseline approach (one of them), and then calculate the arithmetic-mean (AM) for each approach. For example, the relative performance of each approach with respect to c would be a1/c1, b1/c1, c1/c1, a2/c2, b2/c2, and so on. The RATIO-AM with respect to c is now the arithmetic mean of these ratios, i.e., (a1/c1+a2/c2+a3/c3)/3 for a, (b1/c1+b2/c2+b3/c3)/3, and 1 for c. Similarly, RATIO-GM, and RATIO-HM can be obtained by instead calculating geometric mean (GM), or harmonic mean (HM) respectively. Unfortunately, based upon the choice of the baseline approach, the composite ratios can differ (except RATIO-GM, as discussed later). The alternative approach is to calculate means for each approach first, and then find the relative performance ratio. Like before, arithmetic, geometric, or harmonic mean can be used, called AM-RATIO, GM-RATIO, and HM-RATIO respectively. For example, the arithmetic mean of each approach would be (a1+a2+a3)/3 for a, (b1+b2+b3)/3 for b, and so on. The AM-RATIO with respect to c is now the ratio of means of each approach with respect to c’s, i.e., (a1+a2+a3)/(c1+c2+c3) for a, (b1+b2+b3)/(c1+c2+c3) for b, and 1 for c. Since the ratio is calculated last, the choice of the baseline approach has no effect on the result. Note that, GM-RATIO and RATIO-GM give identical composite ratios due to the mathematical property of geometric mean, and thus the order of ratio calculation does not affect the result. Weighted geometric mean has been selected for SPECviewperf® composite numbers [2]. 2. Adjusting Damping factor Adjustment of the damping factor α is a delicate balancing act. For smaller values of α, the convergence is fast, but the link structure of the graph used to determine ranks is less true. Slightly different values for α can produce
  • 3. 3 very different rank vectors. Moreover, as α → 1, convergence slows down drastically, and sensitivity issues begin to surface [1]. For this experiment, the damping factor α (which is usually 0.85) is varied from 0.50 to 1.00 in steps of 0.05. This is in order to compare the performance variation with each damping factor. The calculated error is the L1 norm with respect to default PageRank (α = 0.85). The PageRank algorithm used here is the standard power-iteration (pull) based PageRank. The rank of a vertex in an iteration is calculated as c0 + αΣrn/dn, where c0 is the common teleport contribution, α is the damping factor, rn is the previous rank of vertex with an incoming edge, dn is the out-degree of the incoming-edge vertex, and N is the total number of vertices in the graph. The common teleport contribution c0, calculated as (1-α)/N + αΣrn/N, includes the contribution due to a teleport from any vertex in the graph due to the damping factor (1-α)/N, and teleport from dangling vertices (with no outgoing edges) in the graph αΣrn/N. This is because a random surfer jumps to a random page upon visiting a page with no links, in order to avoid the rank-sink effect. All seventeen graphs used in this experiment are stored in the MatrixMarket (.mtx) file format, and obtained from the SuiteSparse Matrix Collection. These include: web-Stanford, web-BerkStan, web-Google, web-NotreDame, soc-Slashdot0811, soc-Slashdot0902, soc-Epinions1, coAuthorsDBLP, coAuthorsCiteseer, soc-LiveJournal1, coPapersCiteseer, coPapersDBLP, indochina-2004, italy_osm, great-britain_osm, germany_osm, asia_osm. The experiment is implemented in C++, and compiled using GCC 9 with optimization level 3 (-O3). The system used is a Dell PowerEdge R740 Rack server with two Intel Xeon Silver 4116 CPUs @ 2.10GHz, 128GB DIMM DDR4 Synchronous Registered (Buffered) 2666 MHz (8x16GB) DRAM, and running CentOS Linux release 7.9.2009 (Core). The iterations taken with each test case is measured. 500 is the maximum iterations allowed. Statistics of each test case is printed to standard output (stdout), and redirected to a log file, which is then processed with a script to generate a CSV file, with each row representing the details of a single test case. This CSV file is imported into Google Sheets, and necessary tables are set up with the help of the FILTER function to create the charts.
  • 4. 4 Figure 2.1: Geometric mean iterations for PageRank computation with damping factor α adjusted from 0.50 - 1.00 in steps of 0.05. Chart for AM iterations is quite similar. Figure 2.2: Relative geometric mean (GM-RATIO) iterations for PageRank computation with damping factor α adjusted from 0.50 - 1.00 in steps of 0.05. Chart for AM-RATIO is quite similar. Results, as shown in figures 2.1 and 2.2, indicate that increasing the damping factor α beyond 0.85 significantly increases convergence time, and lowering it below 0.85 decreases convergence time. As the damping factor α increases linearly, the iterations needed for PageRank computation
  • 5. 5 increases almost exponentially. On average, using a damping factor α = 0.95 increases iterations needed by 190% (~2.9x), and using a damping factor α = 0.75 decreases it by 41% (~0.6x), compared to damping factor α = 0.85. Note that a higher damping factor implies that a random surfer follows links with higher probability (and jumps to a random page with lower probability). 3. Adjusting Error function It is observed that a number of error functions are in use for checking convergence of PageRank computation. Although L1 norm is commonly used for convergence check, it appears nvGraph uses L2 norm instead [3]. Another person in stackoverflow seems to suggest the use of per-vertex tolerance comparison, which is essentially the L∞ norm [4]. The L1 norm ||E||1 between two (rank) vectors r and s is calculated as Σ|rn - sn|, or as the sum of absolute errors. The L2 norm ||E||2 is calculated as √Σ|rn - sn|2 , or as the square-root of the sum of squared errors (euclidean distance between the two vectors). The L∞ norm ||E||∞ is calculated as max(|rn - sn|), or as the maximum of absolute errors. This experiment was for comparing the performance between PageRank computation with L1, L2 and L∞ norms as convergence check, for damping factor α = 0.85, and tolerance τ = 10-6 . The input graphs, system used, and the rest of the experimental process is similar to that of the first experiment. Additionally, the execution time of each test case is measured using std::chrono::high_performance_timer. This is done 5 times for each test case, and timings are averaged (AM). From the results, as seen in figures 3.1 and 3.2, it is clear that PageRank computation with L∞ norm as convergence check is the fastest, quickly followed by L2 norm, and finally L1 norm. Thus, when comparing two or more approaches for an iterative algorithm, it is important to ensure that all of them use the same error function as convergence check (and the same parameter values). This would help ensure a level ground for a good relative performance comparison. Also note in figures 3.1 and 3.2 that PageRank computation with L∞ norm as convergence check completes in a single iteration for all the road networks (ending with _osm). This is likely because it is calculated as ||E||∞ = max(|rn -
  • 6. 6 sn|), and depending upon the order (number of vertices) N of the graph (those graphs are quite large), the maximum rank change for any single vertex does not exceed the tolerance τ value of 10-6 . Figure 3.1: Iterations needed for PageRank computation for various graphs with damping factor α = 0.85, and tolerance τ = 10-6 , with L1, L2, and L∞ norms as convergence check. Figure 3.2: Relative iterations for PageRank computations for various graphs with L1, L2, and L∞ norms as convergence check, with respect to L1 norm as baseline (damping factor α = 0.85, and tolerance τ = 10-6 ).
  • 7. 7 AM-RATIO 4,645 ms 2,052 ms 1,070 ms 1.00 0.44 0.23 4.34 1.92 1.00 GM-RATIO 1,291 ms 626 ms 236 ms 1.00 0.48 0.18 5.46 2.65 1.00 HM-RATIO 351 ms 198 ms 118 ms 1.00 0.56 0.34 2.98 1.68 1.00 RATIO-AM 1.00 0.50 0.33 15.92 5.66 1.00 RATIO-GM 1.00 0.48 0.18 5.46 2.65 1.00 RATIO-HM 1.00 0.47 0.06 3.03 1.75 1.00 Table 3.1: Mean times for PageRank computation with L1, L2, and L∞ norms as convergence check (first 3 columns), ratios relative to L1 norm (second 3 columns), and then to L∞ norm (last 3 columns). Blue rows are unaffected by baseline choice, purple cells are close to ratios for mean iterations (see table 3.2), and red cells are significantly different. AM-RATIO 57.29 28.82 18.94 1.00 0.50 0.33 3.02 1.52 1.00 GM-RATIO 56.91 27.50 10.21 1.00 0.48 0.18 5.57 2.69 1.00 HM-RATIO 56.51 26.32 3.55 1.00 0.47 0.06 15.91 7.41 1.00 RATIO-AM 1.00 0.50 0.33 17.61 6.08 1.00 RATIO-GM 1.00 0.48 0.18 5.57 2.69 1.00 RATIO-HM 1.00 0.46 0.06 3.00 1.74 1.00 Table 3.2: Mean iterations for PageRank computation with L1, L2, and L∞ norms as convergence check (first 3 columns), ratios relative to L1 norm (second 3 columns), and then to L∞ norm (last 3 columns). Blue rows are unaffected by baseline choice, purple cells are close to ratios for mean iterations (see table 3.1), and red cells are significantly different. In order to obtain a composite relative performance ratio, the six different methods mentioned above are calculated, for both PageRank computation time and iterations. They are listed in table 3.1 and 3.2 respectively. The ratios are calculated with both L1, and L∞ norms as baseline. Methods which calculate ratios at the end are not affected by the choice of baseline, as expected. However, not only is RATIO-GM unaffected as well, it is also identical to GM-RATIO, due to its mathematical property. It is observed that RATIO-AM and RATIO-HM are affected by the choice of baseline, while RATIO-GM is unaffected. RATIO-HM, RATIO-GM, and RATIO-AM with L1 norm as baseline are similar for both time and iterations, but HM-RATIO is significantly different. This indicates GM-RATIO to be the most stable
  • 8. 8 composite relative performance ratio, followed by AM-RATIO. In fact, weighted GM-RATIO is used by SPECviewperf® [2], as mentioned above. Semantically, GM-RATIO comparison gives equal importance to the relative performance of each test case (graph), while an AM-RATIO comparison gives equal importance to magnitude (time/iterations) of all test cases (or simply, it gives higher importance to test cases with larger graphs). 4. Adjusting Tolerance Similar to the damping factor α and the error function used for convergence check, adjusting the value of tolerance τ can have a significant effect. This experiment was for comparing the performance between PageRank computation with L1, L2 and L∞ norms as convergence check, for various tolerance τ values ranging from 10-0 to 10-10 (10-0 , 5×10-0 , 10-1 , 5×10-1 , ...). The input graphs, system used, and the rest of the experimental process is similar to that of the first experiment. For various graphs, some of which are shown in figures 4.1 and 4.2, it is observed that PageRank computation with L1, L2, or L∞ norm as convergence check suffers from sensitivity issues beyond certain (smaller) tolerance τ values, causing the computation to halt at maximum iteration limit (500) without convergence. As tolerance τ is decreased from 10-0 to 10-10 , L1 norm is the first to suffer from this issue, followed by L2 and L∞ norms (except road networks). This sensitivity issue was recognized by the fact that a given approach abruptly takes 500 iterations for the next lower tolerance τ value. It is also observed, as shown in figure 4.2, that PageRank computation with L∞ norm as convergence check completes in just one iteration (even for tolerance τ ≥ 10-6 ) for large graphs (road networks). This again, as mentioned above, is likely because the maximum rank change for any single vertex for L∞ norm, and the sum of squares of total rank change for all vertices, is quite low for such large graphs. Thus, it does not exceed the given tolerance τ value, causing a single iteration convergence.
  • 9. 9 Figure 4.1: Iterations taken for PageRank computation of the web-BerkStan graph, with L1, L2, and L∞ norms used as convergence check. Up till tolerance τ = 5×10-2 , L2 and L∞ norms converge in just a single iteration. From tolerance τ = 10-8 , L1 norm begins to suffer from sensitivity issues, followed by L2 and L∞ norms at 10-9 . Figure 4.2: Iterations taken for PageRank computation of the asia_osm graph, with L1, L2, and L∞ norms used as convergence check. Until tolerance τ = 10-7 , the L∞ norm converges in just one iteration. On average, PageRank computation with L∞ norm as the error function is the fastest, quickly followed by L2 norm, and then L1 norm. This is the case with both geometric mean (GM) and arithmetic mean (AM) comparisons of
  • 10. 10 iterations needed for convergence with each of the three error functions, as shown in figures 4.3 and 4.4. In fact, this trend is observed with each of the individual graphs separately, although not shown here. Figure 4.3: Geometric mean iterations taken for PageRank computation with L1, L2 and L∞ norms as convergence check, and tolerance τ adjusted from 10-0 to 10-10 . Figure 4.4: Arithmetic mean iterations taken for PageRank computation with L1, L2 and L∞ norms as convergence check, and tolerance τ adjusted from 10-0 to 10-10 .
  • 11. 11 Figure 4.5: Relative GM iterations taken for PageRank computation with L1, L2 and L∞ norms as convergence check, and tolerance τ adjusted from 10-0 to 10-10 . Figure 4.6: Relative GM iterations taken for PageRank computation with L1, L2 and L∞ norms as convergence check, and tolerance τ adjusted from 10-0 to 10-10 . Based on GM-RATIO comparison, the relative iterations between PageRank computation with L1, L2, and L∞ norm as convergence check is 1.00 : 0.30 : 0.20. Hence L2 norm is on average 70% faster than L1 norm, and L∞ norm is 33% faster than L2 norm. This ratio is calculated by first finding the GM of iterations based on each error function for each tolerance τ value separately.
  • 12. 12 These tolerance τ specific means are then combined with GM to obtain a single mean value for each error function (norm). The GM-RATIO is then the ratio of each norm with respect to the L∞ norm. The variation of tolerance τ specific means with L∞ norm as baseline for various tolerance τ values is shown in figure 4.5. On the other hand, based on AM-RATIO comparison, the relative iterations between PageRank computation with L1, L2, and L∞ norm as convergence check is 1.00 : 0.39 : 0.31. Hence, L2 norm is on average 61% faster than L1 norm, and L∞ norm is 26% faster than L2 norm. This ratio is calculated in a manner similar to that of GM-RATIO, except that it uses AM instead of GM. The variation of tolerance τ specific means with L∞ norm as baseline for various tolerance τ values is shown in figure 4.6. 5. Conclusion Parameter values can have a significant effect on performance, as seen in these experiments. Different error functions converge at different rates, and which of them converges faster depends upon the tolerance τ value. Iteration count needs to be checked in order to ensure that no approach is suffering from sensitivity issues, or is leading to a single iteration convergence. Finally, the relative performance comparison method affects which results get more importance, and which do not, in the final average. Taking note of each of these points, when comparing iterative algorithms, will thus ensure that the performance results are accurate and useful. The links to source code, along with data sheets and charts, for adjusting damping factor [5], error function [6], and tolerance [7] are included in references. References [1] A. Langville and C. Meyer, “Deeper Inside PageRank,” Internet Math., vol. 1, no. 3, pp. 335–380, Jan. 2004, doi: 10.1080/15427951.2004.10129091. [2] B. Licea-Kane, “SPECviewperf® 7.1 -- Weighted Geometric Mean.”
  • 13. 13 https://www.spec.org/gwpg/gpc.static/geometric.html (accessed Aug. 05, 2021). [3] A. Fender, “nvgraph/pagerank.cu at main · rapidsai/nvgraph.” https://github.com/rapidsai/nvgraph/blob/main/cpp/src/pagerank.cu#L149 (accessed Aug. 06, 2021). [4] N. Upreti, “How to check for Page Rank convergence?” https://stackoverflow.com/a/29321153/1413259 (accessed Aug. 06, 2021). [5] S. Sahu, “puzzlef/pagerank-adjust-damping-factor: Comparing the effect of using different values of damping factor, with PageRank (pull, CSR).” https://github.com/puzzlef/pagerank-adjust-damping-factor (accessed Aug. 06, 2021). [6] S. Sahu, “puzzlef/pagerank-adjust-tolerance-function: Comparing the effect of using different functions for convergence check, with PageRank (pull, CSR).” https://github.com/puzzlef/pagerank-adjust-tolerance-function (accessed Aug. 06, 2021). [7] S. Sahu, “puzzlef/pagerank-adjust-tolerance: Comparing the effect of using different values of tolerance, with PageRank (pull, CSR).” https://github.com/puzzlef/pagerank-adjust-tolerance (accessed Aug. 06, 2021).