Problem 610: Roman Numerals II

(see projecteuler.net/problem=610)

A random generator produces a sequence of symbols drawn from the set {I, V, X, L, C, D, M, #}.
Each item in the sequence is determined by selecting one of these symbols at random, independently of the other items in the sequence.
At each step, the seven letters are equally likely to be selected, with probability 14% each, but the # symbol only has a 2% chance of selection.

We write down the sequence of letters from left to right as they are generated, and we stop at the first occurrence of the # symbol (without writing it).
However, we stipulate that what we have written down must always (when non-empty) be a valid Roman numeral representation in minimal form.
If appending the next letter would contravene this then we simply skip it and try again with the next symbol generated.

Please take careful note of About... Roman Numerals for the definitive rules for this problem on what constitutes a "valid Roman numeral representation" and "minimal form".
For example, the (only) sequence that represents 49 is XLIX. The subtractive combination IL is invalid because of rule (ii), while XXXXIX is valid but not minimal.
The rules do not place any restriction on the number of occurrences of M, so all integers have a valid representation. These are the same rules as were used in Problem 89,
and members are invited to solve that problem first.

Find the expected value of the number represented by what we have written down when we stop. (If nothing is written down then count that as zero.)
Give your answer rounded to 8 places after the decimal point.

My Algorithm

My solution for problem 89 contained two useful functions: number2roman and roman2number.
There is a unique optimal Roman string for each number. The monteCarlo() function generates random string and checks whether x == roman2number(number2roman(x))
But the Monte-Carlo simulation can't find 8 correct digit after the decimal point (it barely finds the correct digits in front of the decimal point).

I split Roman numbers into two groups:

For each number below 1000 I can generate its Roman string (container n2r, read "number-to-Roman") and the inverse mapping r2n ("Roman-to-number").
In a second pass I append to each Roman number each possible Roman letter and check whether this produces a valid, optimal Roman number.

The search() function needs this information to find the expected value of a number (between 0 and 999) and its continuations:
2% * current + 14% * search(next_1) + 14% * search(next_2) + ...
The expected value of all numbers between 0 and 999 can be found by searching all initial letters except M.

Any numbers above 999 consists of one or more M's plus a number between 0 and 999.
I try to add as many M's as possible until the probability of such a number is below my threshold 10^-9

Interactive test

This feature is not available for the current problem.

My code

… was written in C++11 and can be compiled with G++, Clang++, Visual C++. You can download it, too.

#include <iostream>
#include <iomanip>
#include <string>
#include <map>
#include <vector>
#include <cmath>
 
const unsigned int NumLetters = 7;
const char Letters[] = { 'M', 'D', 'C', 'L', 'X', 'V', 'I' };
 
// 2% chance to find the terminal symbol
const double Terminal = 0.02;
// each letter has the same probability of 14% (7*14% + 2% = 100%)
const double AnyLetter = 0.14;
 
// ---------- roman2number and number2roman were copied from problem 89 ----------
 
// convert valid roman numbers to binary numbers
unsigned int roman2number(const std::string& roman)
{
unsigned int result = 0;
 
// remember the value of the previous Roman letter
unsigned int last = 0;
// true, if the current letter is subtracted (and the next identical letters)
bool subtract = false;
 
// walk through the whole string from the end to the beginning ...
for (auto i = roman.rbegin(); i != roman.rend(); i++)
{
unsigned int current = 0;
switch (*i)
{
case 'M': current = 1000; break;
case 'D': current = 500; break;
case 'C': current = 100; break;
case 'L': current = 50; break;
case 'X': current = 10; break;
case 'V': current = 5; break;
case 'I': current = 1; break;
}
 
// smaller than its right neighbor ? => we must subtract
if (current < last)
{
subtract = true;
last = current;
}
// bigger than its right neighbor ? => we must add
else if (current > last)
{
subtract = false;
last = current;
}
 
// note: if current == last then we keep the variables "subtract" and "last" in their current state
 
// add/subtract accordingly
if (subtract)
result -= current;
else
result += current;
}
 
return result;
}
 
std::string number2roman(unsigned int number)
{
// apply these rules in the presented order:
// - as long as number >= steps[i] add roman[i] to result
const unsigned int NumRules = 13;
const unsigned int rules[NumRules] =
{ 1000, 900, 500, 400, 100, 90, 50, 40, 10, 9, 5, 4, 1 };
const char* action[NumRules] =
{ "M", "CM", "D", "CD", "C", "XC", "L", "XL", "X", "IX", "V", "IV", "I" };
 
// apply all rules ...
std::string result;
for (unsigned int i = 0; i < NumRules; i++)
// ... as often as needed
while (number >= rules[i])
{
// reduce integer
number -= rules[i];
// add letter(s)
result += action[i];
}
 
return result;
}
 
// a simple pseudo-random number generator
// (produces the same result no matter what compiler you have - unlike rand() from math.h)
unsigned int myrand()
{
static unsigned long long seed = 0;
seed = 6364136223846793005ULL * seed + 1;
return (unsigned int)(seed >> 30);
}
 
// ---------- problem-specific code ----------
 
// store possible continuations for each Roman representation
std::vector<std::vector<unsigned int>> next(1000);
 
// run randomized simulation
double monteCarlo(unsigned int iterations)
{
double result = 0;
for (unsigned int i = 0; i < iterations; i++)
{
std::string current;
while (true)
{
// choose a random letter
auto choice = myrand() % 1000;
// is it the terminal sign ?
if (choice < Terminal * 1000)
{
result += roman2number(current);
break;
}
 
// append letter
current += Letters[choice % NumLetters];
// still valid ?
auto number = roman2number(current);
auto verify = number2roman(number);
// not valid or not optimal, remove last letter
if (current != verify)
current.pop_back();
}
}
 
return result / iterations;
}
 
// return expected value of current number and all its potential continuations
double search(unsigned int current)
{
// impossible to add any letter and get another valid, optimal number ? => terminal state
if (next[current].empty())
return current;
 
// there's a 2% chance that the current number is the final number
auto result = Terminal * current;
auto percentages = Terminal;
 
// and a 14% chance for each letter 'I', 'V', ...
for (auto x : next[current])
{
result += AnyLetter * search(x);
percentages += AnyLetter;
}
 
// in almost all cases, only a few letters actually yield a valid continuation
// so that those percentages don't add to 100%, need to adjust for that deficit
result /= percentages;
 
return result;
}
 
int main()
{
// store each number's Roman representation (0..999)
std::vector<std::string> n2r(1000);
// map each Roman representation to a number (0..999)
std::map<std::string, unsigned int> r2n;
 
// generate all Roman numbers from 0 to 999
const unsigned int Thousand = 1000;
for (unsigned int i = 0; i < Thousand; i++)
{
auto roman = number2roman(i);
n2r[i] = roman;
r2n[roman] = i;
}
 
// for each Roman number between 0 and 999:
// analyze which numbers they can transition into by adding a single letter (e.g. 50 + 'X' => 60)
for (unsigned int i = 0; i < Thousand; i++)
{
auto current = n2r[i];
 
// add a dummy letter (will be replaced in the next loop)
current += ' ';
 
// try to append every possible letter
for (auto add : Letters)
{
current.back() = add;
// is it a valid number (and optimal) ?
if (r2n.count(current) == 1)
next[i].push_back(r2n[current]);
}
}
 
// process all Roman numbers that DON'T start with an M (it's 0..999)
auto upTo1000 = 0.0;
upTo1000 += AnyLetter * search( 1); // I
upTo1000 += AnyLetter * search( 5); // V
upTo1000 += AnyLetter * search( 10); // X
upTo1000 += AnyLetter * search( 50); // L
upTo1000 += AnyLetter * search(100); // C
upTo1000 += AnyLetter * search(500); // D
 
// stop if adding one more M's doesn't significantly affect the result anymore
auto Precision = 0.000000001; // 10^-9
// count M's => the "thousands"
auto numM = 1;
 
auto result = upTo1000;
while (true)
{
// "weight" of the M's
auto manyM = numM * 1000 * (1 - AnyLetter);
// add the following letters
auto increment = manyM + upTo1000;
// and the whole thing because less likely the more M's we have ...
result += increment * pow(AnyLetter, numM);
 
// enough precision ?
if (increment < Precision)
break;
 
// next iteration
numM++;
}
 
std::cout << std::fixed << std::setprecision(8)
<< result << std::endl;
 
// Monte-Carlo simulation
//while (true)
//std::cout << monteCarlo(100000) << std::endl;
 
return 0;
}

This solution contains 37 empty lines, 52 comments and 6 preprocessor commands.

Benchmark

The correct solution to the original Project Euler problem was found in 0.24 seconds on an Intel® Core™ i7-2600K CPU @ 3.40GHz.
(compiled for x86_64 / Linux, GCC flags: -O3 -march=native -fno-exceptions -fno-rtti -std=gnu++11 -DORIGINAL)

See here for a comparison of all solutions.

Note: interactive tests run on a weaker (=slower) computer. Some interactive tests are compiled without -DORIGINAL.

Changelog

October 9, 2017 submitted solution
October 9, 2017 added comments

Similar problems at Project Euler

Problem 89: Roman numerals

Note: I'm not even close to solving all problems at Project Euler. Chances are that similar problems do exist and I just haven't looked at them.

Heatmap

Please click on a problem's number to open my solution to that problem:

green   solutions solve the original Project Euler problem and have a perfect score of 100% at Hackerrank, too
yellow solutions score less than 100% at Hackerrank (but still solve the original problem easily)
gray problems are already solved but I haven't published my solution yet
blue solutions are relevant for Project Euler only: there wasn't a Hackerrank version of it (at the time I solved it) or it differed too much
orange problems are solved but exceed the time limit of one minute or the memory limit of 256 MByte
red problems are not solved yet but I wrote a simulation to approximate the result or verified at least the given example - usually I sketched a few ideas, too
black problems are solved but access to the solution is blocked for a few days until the next problem is published
[new] the flashing problem is the one I solved most recently
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275
276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450
451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575
576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600
601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616
The 306 solved problems (that's level 12) had an average difficulty of 32.5% at Project Euler and
I scored 13526 points (out of 15700 possible points, top rank was 17 out of ≈60000 in August 2017) at Hackerrank's Project Euler+.

My username at Project Euler is stephanbrumme while it's stbrumme at Hackerrank.

Look at my progress and performance pages to get more details.

more about me can be found on my homepage, especially in my coding blog.
some names mentioned on this site may be trademarks of their respective owners.
thanks to the KaTeX team for their great typesetting library !