def IALS(sp_ratings, alpha=30, iters=10, lamb=0.1, k=10):
    confidence = sp_ratings * alpha
    N_users, N_items = sp_ratings.get_shape()
    X, Y = sp.csr_matrix(np.random.randn(N_users, k)), sp.csr_matrix(np.random.randn(N_items, k))

    I = sp.eye(k) * lamb
    X_I, Y_I = sp.eye(N_users), sp.eye(N_items)

    for i in tqdm(range(iters)):

        Ay = Y.T.dot(Y)
        Ax = X.T.dot(X)

        for u in tqdm(range(N_users)):
            u_row = confidence[u,:].toarray()
            p = u_row.copy()
            p[p != 0] = 1.0

            Cu = sp.diags(u_row, [0])
            B = Y.T.dot(Cu).dot(Y)
            C = Y.T.dot(Cu + Y_I).dot(p.T)
            X[u] = spsolve(Ay + B + I, C)


        for i in tqdm(range(N_items)):
            i_row = confidence[:,i].T.toarray()

            p = i_row.copy()
            p[p != 0] = 1.0

            Ci = sp.diags(i_row, [0]) + X_I
            B = X.T.dot(Ci).dot(X)
            C = X.T.dot(Ci).dot(p.T)
            Y[i] = spsolve(Ax + B + I, C)

    return X, Y

def topn_recommendation_als(sp_ratings, item_vectors, user_vectors, user_id, N=30):
    user_history = sp_ratings[user_id,:].toarray()
    user_history = user_history.reshape(-1) + 1
    user_history[user_history > 1] = 0

    recom = user_vectors[user_id].dot(item_vectors.transpose()).toarray()
    recom = MinMaxScaler().fit_transform(recom.reshape(-1,1))[:,0]
    recommend_vector = user_history * recom

    item_idx = np.argsort(recommend_vector)[::-1][:N]

    topn = pd.DataFrame(item_idx, columns=['itemid'])
    topn['id'] = top_to_bottom_order[topn['itemid']]
    topn = pd.merge(topn, all_names, on='id', how='inner')
    return topn