How to use the class SecureArray?

[1191000814]

How to use the class SecureArray? It is not seem in the document? I want to transfer a np.ndarray type to secure type, how can I do that?

[lschoe]

It's best to look at a demo like demos/np_id3gini.py to see how to work with secure arrays. Also see the other demos listed in demos/np-run-all.sh

For a basic secure type like secint = mpc.SecInt() you can use b = secint.array(a) to convert a Numpy array a into a secret-shared version b, and then work with b using Numpy operations.

[1191000814]

Thanks! I have another problem, the SecInt type is unhashable, but I want to use it as the key of a dict, How can I do?

It's best to look at a demo like demos/np_id3gini.py to see how to work with secure arrays. Also see the other demos listed in demos/np-run-all.sh

For a basic secure type like secint = mpc.SecInt() you can use b = secint.array(a) to convert a Numpy array a into a secret-shared version b, and then work with b using Numpy operations.

[lschoe]

Instances of secure types like secure integers are not hashable because their values are secret-shared. Each party holds a share of the value, but these shares are random and different for each party.

Depending on what you need to do, you might use operators like == and < with secure integers to construct searchable datastructures but this may get quite involved.

[lschoe]

To work with mpc.input() for nested lists, use a similar approach as suggested for mpc.output() above. For instance, use [mpc.input(_, senders=0) for _ in s] if you want only party 0 to provide private input, and s is a list of lists of secure integers, say. Drop the senders=0 argument if you want all parties to have private inputs.

For mpc.if_else(), note that the lists for both branches should be of the same length. You cannot obliviously select between lists of different lengths like you're doing with [1,2] and [1,3,4] because the length of the result would already reveal which branch was followed. Also, make sure to cast the elements of these lists to secure integers, So something like this should work:

b = mpc.if_else(a >= secint(2), [secint(1), secint(2), secint(0)], [secint(1), secint(3), secint(4)])

If you'd work with secure Numpy arrays, there you also have np.where() that can be used with secure integer arrays, as you can see in these test cases for example:

mpyc/tests/test_runtime.py

Lines 120 to 123 in a11ff6a

	np.assertEqual(mpc.run(mpc.output(np.where(c < 0, -c, c))), a)
	np.assertEqual(mpc.run(mpc.output(np.where(c[0] < 0, -c, c))), a)
	np.assertEqual(mpc.run(mpc.output(np.where(True, c, 10))), a)
	np.assertEqual(mpc.run(mpc.output(np.where(False, 10, c))), a)

[1191000814]

Thanks for your patience!
For instance, if I want to add the indexes(not unique) of list my_list which value is a to another list b , How should I do?

[lschoe]

There are several ways to do this. Maybe something like this suits your purposes:

t = [a == _ for _ in my_list]
u = mpc.seclist([-1] * (len(s)+1), secint)
for i in range(len(s), 0, -1):
    u[mpc.sum(t[:i])] = i-1
b = list(u[1:])

[1191000814]

Thanks for you reply!

Maybe there is still a problem, the seclist u contains some other elements because it should ensure its length in the first, so that I will have many troubles in solving the tail elements(I do not know the number of elements, and the seclist can not be range indexed by secint). I want the list u contain no redundant elements.

But... I observed the class secindex, will it be a good idea for the problem? I can not understand anything about secindex until now because of the concise documentation.

[lschoe]

I wouldn't go there. Actually, if you do not need to hide the number of occurrences of a in my_list, then you can first reveal this number, using n = await mpc.output(mpc.sum(t)), and then use n instead of len(s)in what follows. Like this:

t = [a == _ for _ in my_list]
n = await mpc.output(mpc.sum(t))
u = mpc.seclist([-1] * (n+1), secint)
for i in range(len(my_list), 0, -1):
    u[mpc.sum(t[:i])] = i-1
b = list(u[1:])
await mpc.output(b)

[1191000814]

Hi, today when I use await mpc.output() to get the real data of a variable, the program was stuck but not end or error, I don't know why it happens. Is there a dead lock or other reason?

[1191000814]

I used await mpc.out() many times in a program, and it got stuck at one point. It was n = await mpc.output(secure_updated.count(-1)), but if I used a simple clause like n = await mpc.output( secint(3)), it is normal; I'm sorry to bother you, but this problem has puzzled me for a long time.

[1191000814]

By the way, I use -1 as a padding num to make muti-lists the same length before, when I use -2 as the padding num, it can run more long time, when I use 100, it become normal, it' s so strangeness.

All the way, Thanks for your help!