The ContinuousMyhill
transform adapts
values from the driver domain to a bounded range, where the relative concentrations are governed
by John Myhill's enhancement of the exponential distribution.
John Myhill had his own ideas about how Stochastic Music might be generated by computer programs. At the 1978 ICMC, which Xenakis also attended, Myhill proposed "Some Simplifications and Improvements in the Stochastic Music Program". which presentation initially raised some ire on Xenakis's part. An appendix to this presentation took things a bit farther by advancing a model of "controlled indeterminacy"; in which note timings would transition gradually between pulse rhythms, on one hand, and negative exponential randomness (the timings of successive clicks from a Geiger counter), on the other hand.
The range of values output by ContinuousMyhill.convert()
is bounded
below by zero and unbounded above.
The shape of the distribution curve is
controlled by two parameters: scale
, symbolized λ and
ratio
, symbolized ρ. The
scale
must be greater
than zero and the ratio
may not fall below unity. A practical upper bound is indirectly determined
by ratio
.
Each ContinuousMyhill
instance internally maintains a
ContinuousDistribution
instance
which divides the range from zero to this calculated upper bound into trapezoids of equal width. The number of trapezoids is determined
by a fourth parameter maintained as a Java field, itemCount
. For the
present purposes I have set itemCount
to 200.
The quantile function mapping a driver argument to an exponentially distributed result exists in closed form, which allows the present implementation to dispense with all the business of trapezoidal approximation in favor of an in-line mathematical expression:
return -scale * Math.log(driver);
To close things in around the mean, Myhill limited driver values to a subrange from u0
to
u1
, where 0≤
.
The lower bound u0
<u1
≤1u0
is calculated as:
u0 = Math.pow(ratio, 1. / (1. - ratio));
While the upper bound u1
is:
u1
= Math.pow(u0, ratio);
Wikipedia does not recognize the Myhill distribution.
The parametric mean (average, symbolized μ) is identical to the
scale
parameter λ. The variance is 0 when ρ=1 and
converges to 1/λ2 as ρ increases to infinity.
Figure 1 illustrates the influence which
ContinuousMyhill.convert()
exerts over driver sequences when the scale
λ = 1 and
the ratio
ρ = 5. This panel was created using the same driver sources used for the
ContinuousUniform
,
which earlier panel provides a basis for comparison.
ContinuousMyhill
output from three different
Driver
sources. Each row of graphs provides a time-series graph of samples (left)
and a histogram analyzed from the same samples (right).
The first row of graphs was generated using the standard random number generator. The second
row was generated using the balanced-bit generator. The third row was generated using an ascending sequence of driver values,
equally spaced from zero to unity.
The standard-random time-series graph (top row of Figure 1) has the same relative ups and downs as the standard-random time-series graph prepared for
ContinuousUniform
, but the
specific values are squinched down toward zero. Its hard to see how the driver distribution influences the standard-random histogram presented here, other
than the generally ragged shape of the histogram.
The balanced-bit time-series (middle row of Figure 1) likewise has the same ups and downs as the balanced-bit time-series graph prepared for
ContinuousUniform
with
values squinched similarly. Since balanced-bit sequences strive aggressively for uniformity, the jaggedness of this balanced-bit histogram is
accodingly moderated.
The time-series graph generated using ascending, equally spaced driver values (bottom row of Figure 1) presents the percentile function for this flavor of Myhill distribution. The histogram of sample values presents the distribution's probability density function or PDF. The PDF is an equal-ratios curve bending upward from f(v) = 1 when v = 0 to f(v) = 3 when v = 1. Looking back at the time-series graph, notice how the percentile function rises more steeply where the distribution is rarefied and less steeply where the distribution is concentrated.
For each graph in Figure 1 the average sample value is plotted as a
dashed green line, while the interval between ± one standard deviation around
the average is filled in with a lighter green background. For
k = 2 and λ = 5 the parametric average calculates out
to μ = 2×5 = 10, the parametric standard deviation calculates out to
σ = √2×5 = √50 = 7.07, and the upper bound calculates out to
6×7.07 = 42.43. By contrast the numerical
average and deviations for the bottom row of graphs were 10.135 and 6.854. Since this bottommost row illustrates the most ideal conditions
under which a profile can be generated, these parametric and numerical statistics should match closely. I was able to narrow the
gap between parametric and numerical averages by doubling the itemCount
, however this
change did nothing to close the gap between parametric and numerical deviations. I am hoping that the deviation gap will
close with increasing sample counts.
The interval from 10.135-0.280 to 10.135+6.854 is
2*6.854 = 13.708 = 32% of the full application range from zero to unity. Since the continuous uniform distribution
had 58% of samples within ± one standard deviation of the mean, this suggests that the Myhill transform
with scale
1 and ratio
5
is squeezing 58% of samples into 32% of the application range, giving a concentration rate of 58/32 = 1.81.
Figure 2 show how changes in max/min ratios affect the distribution curve. Each figure provides two graphs. The upper graph shows the probability density function or PDF. The lower graph shows the cumulative distribution function or CDF.
When the ratio
parameter ρ is large, the Myhill
distribution simplifies to the exponential distribution that characterizes waiting durations
in the Poisson Point Process.
ContinuousMyhill
implementation class.
The type hierarchy for ContinuousMyhill
is:
TransformBase<T
extends Number>
extends WriteableEntity
implements Transform<T>
ContinuousMyhill
extends TransformBase<Double>
implements Transform.Continuous
Class ContinuousDistributionTransform
embeds a
ContinuousDistribution
instance capable of approximating most any continuous distribution as a succession of trapezoids.
Each ContinuousDistribution
trapezoid item has
left
, right
, origin
,
and goal
fields.
Conversion happens entirely in ContinuousDistributionTransform
,
where the convert()
method does this:
return getDistribution().quantile(driver);
TransformBase
maintains a valid
field
to flag parameter changes. This field starts out false
and reverts to false
with every time ContinuousMyhill
calls TransformBase.invalidate()
. This happens
with any change to shape
, scale
, extent
or itemCount
. Any call to TransformBase.getDistribution()
(and ContinuousDistributionTransform.convert()
makes such a call) first creates
the distribution if it does not already exist, then checks valid
. If false
,
then getDistribution()
calls validate()
, which is
abstract
to TransformBase
but whose implementation is
made concrete by ContinuousMyhill
. And that particular implementation of validate()
makes use of ContinuousDistribution.calculateGamma(shape, scale, extent, itemCount)
to recalculate the
succession of trapezoids.
© Charles Ames | Page created: 2022-08-29 | Last updated: 2022-08-29 |