<s>
In	O
computer	B-General_Concept
science	I-General_Concept
,	O
introselect	B-Algorithm
(	O
short	O
for	O
"	O
introspective	B-Algorithm
selection	I-Algorithm
"	O
)	O
is	O
a	O
selection	B-Algorithm
algorithm	I-Algorithm
that	O
is	O
a	O
hybrid	B-Algorithm
of	O
quickselect	B-Algorithm
and	O
median	B-Algorithm
of	I-Algorithm
medians	I-Algorithm
which	O
has	O
fast	O
average	O
performance	O
and	O
optimal	O
worst-case	O
performance	O
.	O
</s>
<s>
Introselect	B-Algorithm
is	O
related	O
to	O
the	O
introsort	O
sorting	B-Algorithm
algorithm	I-Algorithm
:	O
these	O
are	O
analogous	O
refinements	O
of	O
the	O
basic	O
quickselect	B-Algorithm
and	O
quicksort	B-Algorithm
algorithms	O
,	O
in	O
that	O
they	O
both	O
start	O
with	O
the	O
quick	O
algorithm	O
,	O
which	O
has	O
good	O
average	O
performance	O
and	O
low	O
overhead	O
,	O
but	O
fall	O
back	O
to	O
an	O
optimal	O
worst-case	O
algorithm	O
(	O
with	O
higher	O
overhead	O
)	O
if	O
the	O
quick	O
algorithm	O
does	O
not	O
progress	O
rapidly	O
enough	O
.	O
</s>
<s>
Both	O
algorithms	O
were	O
introduced	O
by	O
David	O
Musser	O
in	O
,	O
with	O
the	O
purpose	O
of	O
providing	O
generic	B-Language
algorithms	I-Language
for	O
the	O
C++	B-Language
Standard	I-Language
Library	I-Language
that	O
have	O
both	O
fast	O
average	O
performance	O
and	O
optimal	O
worst-case	O
performance	O
,	O
thus	O
allowing	O
the	O
performance	O
requirements	O
to	O
be	O
tightened	O
.	O
</s>
<s>
However	O
,	O
in	O
most	O
C++	B-Language
Standard	I-Language
Library	I-Language
implementations	O
,	O
a	O
different	O
"	O
introselect	B-Algorithm
"	O
algorithm	O
is	O
used	O
,	O
which	O
combines	O
quickselect	B-Algorithm
and	O
heapselect	O
,	O
and	O
has	O
a	O
worst-case	O
running	O
time	O
of	O
O(n log n )	O
.	O
</s>
<s>
Introsort	O
achieves	O
practical	O
performance	O
comparable	O
to	O
quicksort	B-Algorithm
while	O
preserving	O
O(n log n )	O
worst-case	O
behavior	O
by	O
creating	O
a	O
hybrid	B-Algorithm
of	O
quicksort	B-Algorithm
and	O
heapsort	B-Application
.	O
</s>
<s>
Introsort	O
starts	O
with	O
quicksort	B-Algorithm
,	O
so	O
it	O
achieves	O
performance	O
similar	O
to	O
quicksort	B-Algorithm
if	O
quicksort	B-Algorithm
works	O
,	O
and	O
falls	O
back	O
to	O
heapsort	B-Application
(	O
which	O
has	O
optimal	O
worst-case	O
performance	O
)	O
if	O
quicksort	B-Algorithm
does	O
not	O
progress	O
quickly	O
enough	O
.	O
</s>
<s>
Similarly	O
,	O
introselect	B-Algorithm
combines	O
quickselect	B-Algorithm
with	O
median	B-Algorithm
of	I-Algorithm
medians	I-Algorithm
to	O
achieve	O
worst-case	O
linear	O
selection	O
with	O
performance	O
similar	O
to	O
quickselect	B-Algorithm
.	O
</s>
<s>
Introselect	B-Algorithm
works	O
by	O
optimistically	O
starting	O
out	O
with	O
quickselect	B-Algorithm
and	O
only	O
switching	O
to	O
a	O
worst-case	O
linear-time	O
selection	B-Algorithm
algorithm	I-Algorithm
(	O
the	O
Blum-Floyd-Pratt-Rivest-Tarjan	O
median	B-Algorithm
of	I-Algorithm
medians	I-Algorithm
algorithm	I-Algorithm
)	O
if	O
it	O
recurses	O
too	O
many	O
times	O
without	O
making	O
sufficient	O
progress	O
.	O
</s>
<s>
The	O
paper	O
suggested	O
that	O
more	O
research	O
on	O
introselect	B-Algorithm
was	O
forthcoming	O
,	O
but	O
the	O
author	O
retired	O
in	O
2007	O
without	O
having	O
published	O
any	O
such	O
further	O
research	O
.	O
</s>
