Unformatted text preview: but
recent
andBoth generalized linear
the of
the to fault havethese yet done this. ofchange a module has analysis employson variant of the sets but requirin
based a larger data
we potential and the illuminate times as the primary are less likely to have predictive CDI EF
not indices number
lines
agentAn alternative (and less powerful) model, using the CDI ªsum of touched file sizesº changes given
creating
though faults doeqi
m arise
not
n are been changed is faults. (EvenX
a tbetter predictorc than its size of the with the effort for individual term in (6) omit
sa
1f e> m absence
;
S
pqvw this is not Â
spontaneously,
m; P a tautology: The g Â of other results are suggestivein [25].)
of (5) faults it will suffer asthe a generalized 0 is the high values, is but, because of the small sam
es of number ofand the same datainPÁ future. That T linear model, is faults. Modules that
c (9), is highly informative.)
terms such as size and complexity
not definitive.
that
nes
g at a primary (and direct) evidence X changes induce faults:
frequentExtreme variability of theafeaturele
changes are
10
The indices, depictchangesÂ deltas c upmsize would
mthe
In (4), :
The taking logarithms of from set of 54
>
where FP0the
entire:017ofoftemporal effects.:64t andbe
Á GLM m set
is past differing thef same to time AGE eqi is model was fit using dataall variables. (Th
gÂ
line
the Where of changes are ªdampedº1andeattenuate over time, As 10
in Section 2.1,
effects
m notedtion, log1 haveavoids negative num
high
c
indistinguishable from one another where , none could
P areore likely to Á, features are the units of
discussed in Section 4.2.4 and and, hence, , I , and
stand while in (5), faults are less likely in older code (provided b is functionality (e.g., call waiting) by which the sy
ted
be posited to have any specific effect.
fhave and is
aults
nged estimated using statistical analysis.
This model implies that code having many lines thatextended modelare too aggregated for most pu
and
The modelthese indices illuminate that some modules
(9) does provide
Both ofchanges are long ratherevidence change as the primary of
10. Here,
survived for a deltas time than MRs.
(person hours)
nes are more decayed than others. Inis likely to be relatively free
principle, this issue could However, effort
dataEFF
c :32 are available
log 1
:13
agent creating precisely, according to faults do a year this level. (Further analysis of factors
log1
faults.
faults. More allowing (Even though (10), code not arise
older
affecting to be module dependent, but
are be addressed by
spontaneously, this is not a tautology:to have only twothirds on larger data sets but À :09
log1 DEL
c
than not
otherwise similar code tends The absence of other
based
requiring the imput
of we have10, 13 yet done this.
Tuesday, September such faults. and complexity is highly informative.) rt for individual changes given aggregated
termsalternative size less powerful) model, using the CDI effo
as
An many as (and be predicted from symptoms and risk factors for decay? The
d the most
analysis employs a variant of
predictive CDI EFF of (6),
odule has estimated to be less than one, the in our data). Statistical 5.1 Temporal Behavior of the Span of C
as
with the ªsum of touched file sizesº term in (6) omitted. The
size of the analyses of the models appear in Section 5.3.
The CDI FILES of (3) measures the difficulty o T 0 is the results are suggestive but, because of the small sample size,
how many code units (files) need to be chang
not
duce faults: 4.2.6 definitive.
Effort
The model was fit using data from a set of 54 features. implement it. An increase in the span of cha
would be One may construct predictors of the effort (personhours)
As noted
are the units of system symptomatic of decay, as discussed in Sectio
none could required to in Section t2.1, features by fitting regression
im (e.g., n
change
Fig. 3 shows that span is increasing for t
functionality plemecall awaiting) by which the system is
equations for and
extended measurementsaggregated fun...
View
Full
Document
This document was uploaded on 03/16/2014 for the course EE 360f at University of Texas.
 Spring '08
 Staff

Click to edit the document details