b89d864ba91813f3ccce46134fd7eec49a7501ed,librosa/util/_nnls.py,,_nnls,#Any#Any#Any#Any#Any#Any#,57

Before Change

        LAt = np.dot(L, A.T)
    else:
        LAt = np.linalg.solve(np.dot(A.T, A) + rho * np.eye(m), A.T)

    LAtB = np.dot(LAt, B)
    LAtApI = np.dot(LAt, A) - np.eye(m)

After Change

    // This puts our initial iterate X into the column space of A
    // so that we have strong (local) convexity
    X = np.linalg.lstsq(A, B)[0]
    Y = np.zeros(X.shape, dtype=A.dtype)
    np.maximum(X, 0.0, Y)
    W = X - Y

    residual = W.copy()

Instances