Too bad that I didn’t see the post in time! I would have loved to participate. I have a mental mechanism to generate arbitrary amounts of random numbers without any external help—except for a small two-digit seed. I tried it on this task and it turns out it is not fast enough. I could only generate 120 bits in ten minutes. I continued the process and here is my sequence:
Thank you Measure. I think that is an OK result for zero preparation but I expected it to do better given that is it based on a small linear congruential generator + noise.
One of the main parts of my scoring system is a program that compares the relative frequency of various substrings of equal length, such as “0000” vs. “0101″. It calculates a score for each string using this method and then looks at what fraction of a large population of truly random strings have a lower score than the sample string. Your first string scored better than 40% of random strings on this metric, and your second string was better than only 1.4% (mostly because it has so many more ones than zeros − 57:93).
Keep in mind that these scores are assuming an equal mix of truly random and user-submitted strings, and the actual guesses depend on how good the other submissions are. If everyone submitted very strongly random strings, all my guesses would be around 50%. (My highest guess on any string was 82% random)
Thank you again. Makes sense that the second one has more zeros—because the original range is 0..48. I did compensate for that in my manual sequence by negating every second and I was very careless (a.k.a. noise) with how I put groups of bits together. I think I could arrive at a more random sequence but it would take much longer to generate.
Too bad that I didn’t see the post in time! I would have loved to participate. I have a mental mechanism to generate arbitrary amounts of random numbers without any external help—except for a small two-digit seed. I tried it on this task and it turns out it is not fast enough. I could only generate 120 bits in ten minutes. I continued the process and here is my sequence:
11010000 11100110 10011101 10101101 01001001 00111011 01001000 10000110
00100011 11000111 11010011 11010000 11110110 11011101 11110000 10110111
10011010 00110100 011111
I’m curious how it compares.
My scoring system would have assigned this string a 68% chance of being truly random, scoring it above 73% of the other strings.
Thank you Measure. I think that is an OK result for zero preparation but I expected it to do better given that is it based on a small linear congruential generator + noise.
How would this pure sequence score?
100001010101110100011111001111001100111000111011010110111110001000011111100110100111011111011111010111000110111010110110111101101100101111101110011101
It was generated like this:
var a = “”; var n = 12;
for(i = 0; i < 38; i++) { n = (n*8+1)%49; a = a + (n%32).toString(2); }
console.log(a);
That one scores as only 58% random.
One of the main parts of my scoring system is a program that compares the relative frequency of various substrings of equal length, such as “0000” vs. “0101″. It calculates a score for each string using this method and then looks at what fraction of a large population of truly random strings have a lower score than the sample string. Your first string scored better than 40% of random strings on this metric, and your second string was better than only 1.4% (mostly because it has so many more ones than zeros − 57:93).
Keep in mind that these scores are assuming an equal mix of truly random and user-submitted strings, and the actual guesses depend on how good the other submissions are. If everyone submitted very strongly random strings, all my guesses would be around 50%. (My highest guess on any string was 82% random)
Thank you again. Makes sense that the second one has more zeros—because the original range is 0..48. I did compensate for that in my manual sequence by negating every second and I was very careless (a.k.a. noise) with how I put groups of bits together. I think I could arrive at a more random sequence but it would take much longer to generate.