Python has the ability to create random class instances, each a separate RNG (Random Number Generator)

This allows multiple independent random number generators to be created and used that do not share state.

This is done with the random.Random class.

This can be useful for certain simulations whereby certain types of repeatable randomness is needed in the presence of parts of a program that change.

Here is the Python code [#1]
import random def randTestDo1(rlen1, clen1, order1): print("Randoms: {0:d} x {1:d} by {2:s}".format(rlen1,clen1,order1)) # shared RNG to show difference rand0 = random.Random(0) spos1 = 0 randDict1 = {} seedList1 = [] for rpos1 in range(1,rlen1+1): for cpos1 in range(1,clen1+1): seed1 = (rpos1-1)*clen1+cpos1 randDict1[(rpos1,cpos1)] = random.Random(seed1) s1 = "({0:d},{1:d}):{2:d}".format(rpos1,cpos1,seed1) seedList1.append(s1) s1 = " ".join(seedList1) print("Seeds: {0:s}".format(s1)) if order1 == "row": print("Row major order:") for rpos1 in range(1,rlen1+1): for cpos1 in range(1,clen1+1): rand1 = randDict1[(rpos1,cpos1)] r0 = rand0.random() r1 = rand1.random() print("\t\t({0:d},{1:d}) {2:0.4f} {3:0.4f}".format(rpos1,cpos1,r1,r0)) else: print("Column major order:") for cpos1 in range(1,clen1+1): for rpos1 in range(1,rlen1+1): rand1 = randDict1[(rpos1,cpos1)] r0 = rand0.random() r1 = rand1.random() print("\t\t({0:d},{1:d}) {2:0.4f} {3:0.4f}".format(rpos1,cpos1,r1,r0)) randTestDo1(2,2,"row") print("") randTestDo1(2,2,"col")
Here is the output of the Python code.
Randoms: 2 x 2 by row Seeds: (1,1):1 (1,2):2 (2,1):3 (2,2):4 Row major order: (1,1) 0.1344 0.8444 (1,2) 0.9560 0.7580 (2,1) 0.2380 0.4206 (2,2) 0.2360 0.2589 Randoms: 2 x 2 by col Seeds: (1,1):1 (1,2):2 (2,1):3 (2,2):4 Column major order: (1,1) 0.1344 0.8444 (2,1) 0.2380 0.7580 (1,2) 0.9560 0.4206 (2,2) 0.2360 0.2589
The random numbers for the lower-left (2, 1) cell and the upper-right (1, 2) cell are different for the shared RNG but the same when using and independent RNG for each cell.

One way to represent a 2D table structure is as follows, using lists of lists.

There is no easy or convenient way to store meta information about the table in the list structure.

For many applications, where appropriate, I use the following approach for a 2D table that uses a dictionary.

The tuples represent the cells, but other meta information can be used.

Note: A class can be used to load, save, etc., but this structure makes the JSON save/load convenient.

In same cases, the list structure can become one dictionary key, if needed.