utils.py 文件源码

def check_embed_match(X_embed1, X_embed2):
    """
    Check whether the two embeddings are almost the same by computing their normalized euclidean distances
    in the embedding space and checking the correlation.
    Inputs:
        - X_embed1, X_embed2: two Nxd matrices with coordinates in the embedding space
    Returns:
        - r: Pearson correlation coefficient between the normalized distances of the points
    """
    D_emb1 = pdist(X_embed1, 'euclidean')
    D_emb2 = pdist(X_embed2, 'euclidean')
    D_emb1 /= D_emb1.max()
    D_emb2 /= D_emb2.max()
    return np.corrcoef(D_emb1, D_emb2)[0, 1]