Rescale to next funcion

[safiia]

Description
When I used rescale_to_next(), it did not work as it should. It turns cypher text from level 1, I mean <mod_level=1>, to none instead of level 0 <mod_level=0>

Code To Reproduce Error

    def forward(self, encrypted_indices, relin_key):
        print(f"Forward pass of EncryptedEmbedding with input length: {len(encrypted_indices)}")
        result = []
        for enc_index in encrypted_indices:
            if enc_index is None:
                print("Warning: enc_index is None. Skipping.")
                continue

            one_hot_vector = [self.HE.encryptFrac(np.array([0.0], dtype=np.float64)) for _ in range(len(self.embedding_matrix))]
            for i in range(len(self.embedding_matrix)):
                enc_i = self.HE.encryptFrac(np.array([float(i)], dtype=np.float64))
                print("enc_i",enc_i)
                print("enc_index",enc_index)
                comparison_result = self.HE.square(self.HE.sub(enc_index, enc_i))
                print("comparison_result", comparison_result)
                comparison_result = self.HE.rescale_to_next(comparison_result)
                print("comparison_result", comparison_result)
                \`\`\`one_hot_element = self.HE.sub(self.HE.encryptFrac(np.array([1.0], dtype=np.float64)), comparison_result)\`\`\`
                print("one_hot_element", one_hot_element)

                for j in range(len(self.embedding_matrix)):
                    print("i=", i, "j=", j)
                    one_hot_element, one_hot_vec_elem = self.align_scales(one_hot_element, self.one_hot_vectors[i][j])
                    # print("one_hot_element", one_hot_element)
                    # print("one_hot_vec_elem", one_hot_vec_elem)
                    multiplied_element = self.custom_rescale_to_next(self.HE.multiply(one_hot_element, one_hot_vec_elem))
                    print("multiplied_element", multiplied_element)
                    one_hot_vector[j] = self.HE.add(one_hot_vector[j], multiplied_element)
            print(f"Size of one_hot_vector: {len(one_hot_vector)}")
            for idx, vec in enumerate(one_hot_vector):
                print(f"one_hot_vector[{idx}]: scale={vec.scale}, mod_level={vec.mod_level}")

            embedded_vector = self.encrypted_matrix_vector_multiplication(self.HE, self.embedding_matrix, one_hot_vector, relin_key)
            result.append(embedded_vector)
        return result

Result:
enc_i <Pyfhel Ciphertext at 0x7b2ba1e1f0b0, scheme=ckks, size=2/2, scale_bits=30, mod_level=0>
enc_index <Pyfhel Ciphertext at 0x7b2fc82e5d50, scheme=ckks, size=2/2, scale_bits=30, mod_level=0>
comparison_result <Pyfhel Ciphertext at 0x7b2fc82e5d50, scheme=ckks, size=3/3, scale_bits=60, mod_level=1>
comparison_result None

Pyfhel/Pyfhel.pyx in Pyfhel.Pyfhel.Pyfhel.sub()

RuntimeError: scheme type mistmatch in sub terms ({ctxt._scheme} VS {ctxt_other._scheme}

I expect the value of comparison_result to be rescaled to (level 0) instead of none, but I think the rescale function behaviour.

My Setup:

• OS: [Ubuntu]
• Python: 3.8
• C compiler version: (AMD64)
• Pyfhel Version: 3.4.2

[ShokofehVS]

I was reading some issues having somewhat similar descriptions. Should it be due to runnig out of available rescalings? Then the problem can be solved following #224.