TEST: monobit_test
Ones count = 99539996
Zeroes count = 99677532
FAIL
P=1.9503405527150787e-22

success = (p >= 0.1) return (success,p,None)

Monobit test checks against 0.1. It should be 0.01 right ?

Another failure I see is if you pass same values as test. For example
bits = np.random.randint(1, size=1000)

Traceback (most recent call last):
File "randomness_check.py", line 103, in
success,p,plist = func(bits)
File "/Users/karthik/sp800_22_tests/sp800_22_approximate_entropy_test.py", line 73, in approximate_entropy_test
sum += Ci[i]*math.log((Ci[i]/10.0))
ValueError: math domain error

I have question about random bit and random number :
I know random bit is necessary for random number, but I dont know if random number is ok random bit is ok? I want to test random bit generator in Linux.

bits = np.random.randint(2, size=120)

For small dataset, longest_run_ones_in_a_block_test doesn't fail correctly and causes run to stop

`def longest_run_ones_in_a_block_test(bits):
n = len(bits)

if n < 128:
    #return 0 [Current]
    return (None,None,None)`

Hi,
any way to run with python2.6?! I use Python 2.6.6 on centos 6

./sp800_22_tests.py /tmp/megabitrand.bin
Tests of Distinguishability from Random
TEST: monobit_test
Ones count = 523749
Zeroes count = 524827
Traceback (most recent call last):
File "./sp800_22_tests.py", line 132, in
(success,p,plist) = func(bits)
File "/root/git/sp800_22_tests/sp800_22_monobit_test.py", line 44, in monobit_test
p = math.erfc(float(s)/(math.sqrt(float(n)) * math.sqrt(2.0)))
AttributeError: 'module' object has no attribute 'erfc'

Hi,

I tried to do the NIST randomnes test and have downloaded the source file for both sp test and djenrandom, and compiled it using the makefile. When I type the command:
./djenrandom -b -k 128 > megabitrand.bin

The system outputs an error:

Failed to open YAML output file for writing
: No such file or directory

I am not sure if there any thing worng. Could you give me any advice? Thanks

This is similar to issue #10 , but for tests sp800_22_linear_complexity_test.py and sp800_22_maurers_universal_test.py.

Right now, those 2 tests terminates (exit()) the running script if the dataset is too small. As fixed on issue #10, they would work best if just return a fail test (return False), so it doesn't terminate the python script sp800_22_tests.py.

Hi! Is the code specifically defined for 8-bit format, it shouldn't work for less bits?

in sp800_22_overlapping_template_matching_test.py
line 52 : M = 1062
line 54 : print("Insufficient data. %d bit provided. 1,028,016 bits required" % len(bits))
but in NIST SP 800-22 Revision 1a, 2.8.2 Function Call :
M The length in bits of a substring of ε to be tested. M has been set to 1032 in the test code.

So I think it should be :
line 52 : M = 1032
line 54 : print("Insufficient data. %d bit provided. 998,976 bits required" % len(bits))

Hi dj, Thank you for converting sp800-22 to python code。
I also implement DFT test using python ,and get the same p_value as yours. But I am confused is why the test results of bits "1001010011" inside the sp800-22ra1 document is 0.029523, our implement get 0.818545808382.

Thanks.

See pg. 2-39 of the reference material:

https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-22r1a.pdf

Basically it seems that what's currently calculated as p should be passed into an erfc function, and then THAT will be the actual p.

hi

Thank you for converting sp800-22 to python code.

I am using this program very usefully.

The p-value criterion of sp800_22_monobit_test.py is set to 0.1.

Is not 0.01 correct?

SUMMARY

monobit_test 0.581114294834 PASS
frequency_within_block_test 0.747793618708 PASS
runs_test 0.842033492763 PASS
longest_run_ones_in_a_block_test 0.863705090529 PASS
binary_matrix_rank_test 0.0359437277529 FAIL

in the source code:

p = math.e **(-chisq/2.0)
success = (p >= 0.1)

why 0.1，i think maybe 0.01？

Hey,
I am running in to a problem when trying to execute the randomness tests.
My Setup: 16GB RAM; Fedora 29; random-bits-file is about 100MB in size; swap-partion: 8GB; Python 2.7.15 installed; Python 3.7.2 installed
Reproducing steps in Fedora

• Start the test with pyhton3 sp800_22_tests.py random-bit-file
  Test runs with about 40% RAM allocated. When coming to the DFT-test the allocation increases to the maximum. See sp800-22_memorry_error.txt and sp800-22_memorry_error2.txt for details.

• Note that calling the tests with the "old" python 2.7 results in a simple Killed message during the runs_test: sp800-22_killed.txt

I've created 1 megabyte of zeros as dd if=/dev/zero of=zeros.bin bs=1024 count=1024

and when run sp800_22_tests.py on this file it fails like that:

Traceback (most recent call last):
  File "sp800_22_tests.py", line 139, in <module>
    (success,p,plist) = func(bits)
  File "/Users/catap/src/sp800_22_tests/sp800_22_frequency_within_block_test.py", line 74, in frequency_within_block_test
    p = gammaincc((num_of_blocks/2.0),float(chisq)/2.0)
  File "/Users/catap/src/sp800_22_tests/gamma_functions.py", line 67, in gammaincc
    return upper_incomplete_gamma(a,x)/gamma(a)
  File "/Users/catap/src/sp800_22_tests/gamma_functions.py", line 16, in upper_incomplete_gamma
    result = ((x**a) * (e**(-x)))/upper_incomplete_gamma(a,x,d=d+1)
OverflowError: (34, 'Result too large')

Hi ! Thank you for sharing your programs on GitHub ! It's a useful tool.
In the first, I suggest that sp800_22_tests programs be indicated to use in Python 2.7 in a prominent position of the web page on GitHub. They can't be run in Python 2.6. After a slight transform, they can be run in Python 3. The steps of transform are:
Transform 'print' to the function form, that is, 'print( )' in all python files ;
Transform 'xrange' to 'range' in all python files;
In file 'sp800_22_dft_test.py', Line 37: mags = abs(fs)[:n/2] # Compute magnitudes of first half of sequence
Transform 'n/2' to 'n//2', that is, mags = abs(fs)[:n//2]
In file 'sp800_22_tests.py', Line 36: byte = ord(bytech)
Transform 'ord(bytech)' to 'bytech', that is, byte = bytech
After finishing the above modification, I can run the programs successfully in Python 3.6.5.
Next, NIST provides a tool on web site: https://csrc.nist.gov/CSRC/media/Projects/Random-Bit-Generation/documents/sts-2_1_2.zip
In the zip file there are some random number samples. Especially, in \sts-2.1.2\data there are three binary data sample files: BBS.dat, data_bad_rng, data.sha1. I tested the three files by python programs. MemoryError was reported while testing BBS.data file, maybe the file size is too big. The file data.sha1 can pass the test. To my surprise, the file data.bad_rng can pass the test. In my opinion, data.bad_rng is a negative example to demonstrate bad randomness. I wonder why such a bad random number collection sample given by NIST can pass the test. Thanks a lot !

In the file 'sp800_22_tests.py' :

def read_bits_from_file(filename,bigendian):
bitlist = list()
if filename == None:
f = sys.stdin
else:
f = open(filename, "rb")

A file seems to be opened, but I cannot find the place where it is closed. Is this a small bug?