dyce provides two core primitives for enumeration1.
1
>>>fromdyceimportH,P
H objects represent histograms for modeling discrete outcomes.
They encode finite discrete probability distributions as integer counts without any denominator.
P objects represent pools (ordered sequences) of histograms.
If all you need is to aggregate outcomes (sums) from rolling a bunch of dice (or perform calculations on aggregate outcomes), H objects are probably sufficient.
If you need to select certain histograms from a group prior to computing aggregate outcomes (e.g., taking the highest and lowest of each possible roll of n dice), that’s where P objects come in.
As a wise person whose name has been lost to history once said: “Language is imperfect. If at all possible, shut up and point.”
So with that illuminating (or perhaps impenetrable) introduction out of the way, let’s dive into some examples!
Basic examples
A six-sided die can be modeled as:
12
>>>H(6)H(6)
H(n) is shorthand for explicitly enumerating outcomes \([{{1} .. {n}}]\), each with a frequency of 1.
12
>>>H(6)==H({1:1,2:1,3:1,4:1,5:1,6:1})True
Tuples with repeating outcomes are accumulated.
A six-sided “2, 3, 3, 4, 4, 5” die can be modeled as:
12
>>>H((2,3,3,4,4,5))H({2:1,3:2,4:2,5:1})
A fudge die can be modeled as:
12
>>>H((-1,0,1))H({-1:1,0:1,1:1})
Python’s matrix multiplication operator (@) is used to express the number of a particular die (roughly equivalent to the “d” operator in common notations). The outcomes of rolling two six-sided dice (2d6) are:
Where n is an integer, P(n,...) is shorthand for P(H(n),...).
Python’s matrix multiplication operator (@) can also be used with pools.
The above can be expressed more succinctly.
12
>>>2@P(6)P(6,6)
Pools (in this case, Sicherman dice) can be compared to histograms.
Histograms should be sufficient for most calculations.
However, pools are useful for “taking” (selecting) only some of each roll’s outcomes.
This is done by providing one or more index arguments to the P.h method or the P.rolls_with_counts method.
Indexes can be integers, slices, or a mix thereof.
Outcome indexes are ordered from least to greatest with negative values counting from the right, as one would expect (i.e., [0], [1], …, [-2], [-1]).
Summing the least two faces when rolling three six-sided dice would be:
1234
>>>3@P(6)P(6,6,6)>>>(3@P(6)).h(0,1)# see warning below about parenthesesH({2:16,3:27,4:34,5:36,6:34,7:27,8:19,9:12,10:7,11:3,12:1})
Mind your parentheses
Parentheses are needed in the above example because @ has a lower precedence than . and […].
123456
>>>2@P(6).h(1)# equivalent to 2@(P(6).h(1))Traceback(mostrecentcalllast):...IndexError:tupleindexoutofrange>>>(2@P(6)).h(1)H({1:1,2:3,3:5,4:7,5:9,6:11})
Taking the least, middle, or greatest face when rolling three six-sided dice would be:
>>>d10=H(10)-1;d10# a common “d10” with faces [0 .. 9]H({0:1,1:1,2:1,3:1,4:1,5:1,6:1,7:1,8:1,9:1})>>>h=P(4,6,8,d10,12,20).h(0,-1)>>>print(h.format(scaled=True))avg|13.48std|4.40var|19.391|0.00%|2|0.01%|3|0.06%|4|0.30%|#5|0.92%|#####6|2.03%|###########7|3.76%|####################8|5.57%|##############################9|7.78%|###########################################10|8.99%|##################################################11|8.47%|###############################################12|8.64%|################################################13|8.66%|################################################14|6.64%|####################################15|5.62%|###############################16|5.16%|############################17|5.00%|###########################18|5.00%|###########################19|5.00%|###########################20|5.00%|###########################21|4.50%|#########################22|2.01%|###########23|0.73%|####24|0.18%|
Both histograms and pools support various comparison operations as well as substitution.
The odds of observing all even faces when rolling \(n\) six-sided dice, for \(n\) in \([1..6]\) is:
The odds of scoring at least one nine or higher on any single die when rolling \(n\) “exploding” six-sided dice, for \(n\) in \([1..10]\) is:
1 2 3 4 5 6 7 8 91011121314151617
>>># By the time we're rolling a third die, we're guaranteed a nine or higher, so we only need to look that far>>>exploding_d6=H(6).explode(max_depth=2)>>>forninrange(10,0,-1):...d6e_ge_9=exploding_d6.ge(9)...number_of_nines_or_higher_in_nd6e=n@d6e_ge_9...at_least_one_9=number_of_nines_or_higher_in_nd6e.ge(1)...print(f"{n: >2}d6-exploding: {at_least_one_9[1]/sum(at_least_one_9.counts()): >6.2%}")10d6-exploding:69.21%9d6-exploding:65.36%8d6-exploding:61.03%7d6-exploding:56.15%6d6-exploding:50.67%5d6-exploding:44.51%4d6-exploding:37.57%3d6-exploding:29.77%2d6-exploding:20.99%1d6-exploding:11.11%
Dependent probabilities
In many cases, dependent probabilities can be compactly expressed via H.substitute or resolve_dependent_probability, although it takes some practice to adjust to the semantics.
Where we can identify independent terms and reduce dependent terms to calculations solely involving independent terms, there is a fairly straightforward translation.
First, we express independent terms as histograms.
Second, we express dependent terms as substitution functions.
Finally, we pass the dependent substitution functions to the independent histograms via H.substitute or resolve_dependent_probability.
Say we want to roll a d6 and compare whether the result is strictly greater than its distance from some constant.
Now, instead of a constant, let’s use another die as a second independent variable.
We’ll roll a d4 and a d6 and compare whether the d6 is strictly greater than the absolute difference between dice.
12
>>>d4=H(4)# first independent term>>>d6=H(6)# second independent term
One way to express the dependent terms is to nest substitution functions, where the innermost holds the dependent probability calculation, and the outer functions each establish the scope of their respective independent outcomes.
The outer ring and corresponding labels can be overridden for interesting, at-a-glance displays.
Overrides apply counter-clockwise, starting from the 12 o’clock position.
The outer ring can also be used to compare two histograms directly.
Ever been curious how your four shiny new fudge dice stack up against your trusty ol’ double six-siders?
Well wonder no more!
The dyce abides.
Thanks to numerary, dyce offers best-effort support for arbitrary number-like outcomes, including primitives from symbolic expression packages such as SymPy.
Be aware that, depending on implementation, performance can suffer quite a bit when using symbolic primitives.
For histograms and pools, dyce remains opinionated about ordering.
For non-critical contexts where relative values are indeterminate, dyce will attempt a “natural” ordering based on the string representation of each outcome.
This is to accommodate symbolic expressions whose relative values are often unknowable.
SymPy does not even attempt simple relative comparisons between symbolic expressions, even where they are unambiguously resolvable.
Instead, it relies on the caller to invoke its proprietary solver APIs.
dyce, of course, is happily ignorant of all that keenness.
(As it should be.)
In practice, that means that certain operations won’t work with symbolic expressions where correctness depends on ordering outcomes according to relative value (e.g., dice selection from pools).
dyce also provides additional primitives (R objects and their kin) which are useful for producing weighted randomized rolls without the overhead of enumeration.
These are covered seperately. ↩