def cold_start_similar_items(feat_idxs, item_feat_mtx, model, N=10:
feat_mat = scipy.sparse.coo_matrix((np.ones_like(feat_idxs),
(feat_idxs, np.zeros_like(feat_idxs))))
repr, bias = model.item_embeddings(feat_mat)
scores = item_representation.dot(repr[0].T)
# snip