encrypted1 and encrypted2 parameter mismatch
Propantheline opened this issue · 6 comments
Description
The encrypted 564 matrix is multiplied with 64784 plaintext matrix by the mm() function, and it reports ValueError: encrypted1 and encrypted2 parameter mismatch.
Confusion
The encrypted matrix is multiplyable with np.random.rand(64,784), and mnist's data(64*784) will report an error, why is that?
Hi @Propantheline, I came across similar error, here is my code
import tenseal as ts
ctx = ts.context(ts.SCHEME_TYPE.CKKS,
poly_modulus_degree=8192,
coeff_mod_bit_sizes=[35,27,27,35])
ctx.global_scale = pow(2, 27)
ctx.generate_galois_keys()
x = [[0.1, 0.1], [0.1, 0.0]]
# x = [[0.1, 0.1], [0.1, 0.1]]
w = ts.ckks_tensor(ctx, [[0.1, 0.1], [0.1, 0.1]])
z = w.mm(x)
d = z.mm(x)When x = [[0.1, 0.1], [0.1, 0.0]], I got the error below (no error when x = [[0.1, 0.1], [0.1, 0.1]])
Traceback (most recent call last):
File "ten.py", line 14, in <module>
d = z.mm(x)
File "/Users/xxf/miniconda3/envs/ph/lib/python3.8/site-packages/tenseal/tensors/ckkstensor.py", line 118, in mm
result = self.data.mm(other)
ValueError: encrypted1 and encrypted2 parameter mismatch
It seems that it's 0.0 cause this error since MNIST data also contains zeros.
Maybe you could check the shapes of x, w, and z, or you could increase the number of coefficients in coeff_mod_bit_sizes, which determines the number of homomorphic multiplications performed.
Maybe you could check the shapes of x, w, and z, or you could increase the number of coefficients in coeff_mod_bit_sizes, which determines the number of homomorphic multiplications performed.
The shape of x, w, z are all (2,2). And I have try different ckks params, such as increase the number of coefficients, but I still got this error.
Description
The encrypted 5_64 matrix is multiplied with 64_784 plaintext matrix by the mm() function, and it reports ValueError: encrypted1 and encrypted2 parameter mismatch.
Confusion
The encrypted matrix is multiplyable with np.random.rand(64,784), and mnist's data(64*784) will report an error, why is that?
BTW, did you solve this error?
Sorry, I forgot how I did it then, and it didn't seem to fix it, I encrypted the plaintext matrix again and multiplied it with the ciphertext matrix to get no error. What I understand is to keep the ciphertext scale on a level, or the parameters don't match.
Sorry, I forgot how I did it then, and it didn't seem to fix it, I encrypted the plaintext matrix again and multiplied it with the ciphertext matrix to get no error. What I understand is to keep the ciphertext scale on a level, or the parameters don't match.
Thanks, it seems that TenSEAL does relin, rescale, and mod switch automatically. I still wonder why I got this error. Anyway, here is another script which receives a parameter mismatch error.
import tenseal as ts
ctx = ts.context(ts.SCHEME_TYPE.CKKS,
poly_modulus_degree=32768,
coeff_mod_bit_sizes=[60,40,40,40,60])
ctx.global_scale = pow(2, 40)
ctx.generate_galois_keys()
x = [[0.1, 0.1], [0.1, 0.0]]
w = ts.ckks_tensor(ctx, [[0.1, 0.1], [0.1, 0.1]])
z = w.mm(x)
t = [[1., 0.], [0., 1.]]
# t = [[1., 0.0001], [0.0001, 1.]]
d = z * t
d.sum()If t = [[1., 0.], [0., 1.]], I will get an error. But if t = [[1., 0.0001], [0.0001, 1.]], it's all good. The only difference is the first t contains 0..
Traceback (most recent call last):
File "tt.py", line 16, in <module>
d.sum()
File "/Users/xxf/miniconda3/envs/ph/lib/python3.8/site-packages/tenseal/tensors/abstract_tensor.py", line 153, in sum
return self._wrap(self.data.sum(axis))
ValueError: encrypted1 and encrypted2 parameter mismatch