The README states: Note: Cracker does not implement prediction of random() function since it is based on the os.urandom module which is based on /dev/urandom - however, at least in current versions of CPython, this is not true. You can find the implementation of this function for Python 3.10 here: https://github.com/python/cpython/blob/60adc4b92a8a6fe115a023c8f639a6de4730fac1/Modules/_randommodule.c#L153-L177
It's basically these two lines:

uint32_t a=genrand_uint32(self)>>5, b=genrand_uint32(self)>>6;
return PyFloat_FromDouble((a*67108864.0+b)*(1.0/9007199254740992.0));

with genrand_uint32 being the MT random generator, the same as in getrandbits(). One needs to use SystemRandom to get the randomness from the os directly.

There is a slight loss of information about the state here, as in total 11 bits are zeroed, so I'm not sure if it can be used to crack the generator anyway, but it might be a good idea to correct the README at least, and I'll try to add a prediction generator at least.

Mersenne Twister is a determenistic algorithm, in particular in CPython's implementation. Something must be wrong in the code. I ask for help in finding the problem in randcrack's code.

Hello,

I am a student and I dont understand one thing. My project was to research and compare two random number generators. I chose Java's LCG and python's mersenne twister.
Could you explain to me why your (python) RNG prediction loses accuracy over time. I have researched more cracks on Mersenne twister, and they explained that if you have the first 624 integers, you are able to find the seed, as that is where the seed number possibility ends. Is python somehow buffering its seed every step after 624, or what is the issue?

According to the README.md,

Accuracy

Cracker is not absolutely accurate. It is able to perform close to 100% accurate on first 624 32-bit generations, ~99.5% on the first 1 000, ~95% on the first 10 000 and then figures drop to ~50% accurate to generation 50 000.

but I think this is the value of total (or accumulated) accuracy.

When I plot the accuracy of recent 100 predictions, the plot drops much faster than the total accuracy. And IMHO this is a more appropriate measure when discussing prediction accuracy.

Though I do not think what is written in README.md is wrong, I think it is a bit confusive.

.

import random, time
from randcrack import RandCrack

random.seed(10)

rc = RandCrack()

for i in range(624):
	rc.submit(random.randrange(0,5))
	# Could be filled with random.randint(0,4294967294) or random.randrange(0,4294967294)

for i in range(20):
	print("Random result: {}\nCracker result: {}".format(random.randrange(0,5), rc.predict_randrange(0,5)))

Random result: 4
Cracker result: 2
Random result: 4
Cracker result: 4
Random result: 0
Cracker result: 2
Random result: 3
Cracker result: 0
Random result: 2
Cracker result: 0
Random result: 3
Cracker result: 4
Random result: 1
Cracker result: 0
Random result: 0
Cracker result: 0
Random result: 3
Cracker result: 0
Random result: 3
Cracker result: 0
Random result: 0
Cracker result: 4
Random result: 0
Cracker result: 2
Random result: 3
Cracker result: 0
Random result: 1
Cracker result: 0
Random result: 0
Cracker result: 0
Random result: 3
Cracker result: 0
Random result: 4
Cracker result: 4
Random result: 0
Cracker result: 0
Random result: 0
Cracker result: 4
Random result: 4
Cracker result: 2
Press any key to continue . . .

Only 25% accuracy.

from randcrack import RandCrack

ModuleNotFoundError Traceback (most recent call last)
in
----> 1 from randcrack import randcrack

ModuleNotFoundError: No module named 'randcrack'

from randcrack import RandCrack
this code is not working after installing randcrack using pip on python 3.6. using jupyter notebook.
can you please help me out?

Hello,
I noticed that predictor doesn't work on getrandbits when we request too many requests. You can see that using

#!/bin/env python3

import random, time
from randcrack import RandCrack

random.seed(time.time())

rc = RandCrack()

for i in range(624):
	rc.submit(random.getrandbits(32))
	# Could be filled with random.randint(0,4294967294) or random.randrange(0,4294967294)

for i in range(15000):
    if random.getrandbits(32) != rc.predict_getrandbits(32):
        print("not ok, i = ", i)

I installed using pip3 so I've the last version.