From: Matrix factorization with neural network for predicting circRNA-RBP interactions
Algorithm 2: The MFNN with P-U learning Algorithm | |
Input:Y: the known interaction matrix, T: the times of sampling round Set: Obtain set P and U from Y, K: the size of P in each sampling round | |
Output:Fu: unlabeled sample score | |
Step 1: Initialize ∀u ∈ U, t(u) ← 0, MFNN(u) ← 0 | |
Step 2: For t from 1 to T do | |
Randomly sample the set Ut of size K in U. | |
Train a model MFNNt to discriminate P against Ut | |
For ∀u ∈ U\Ut, update: | |
MFNN(u) ← MFNN(u) + MFNNt(u) | |
t (u) ← t(u) + 1 | |
end For | |
Step 3: Return Fu = MFNN(u)/t(u) for u ∈ U |