The reproduction results of the DLPFC dataset are inconsistent with those recorded in the paper

[JingofGao]

My code:

samples = ["151507", "151508", "151509",
        "151510",` "151669", "151670",
        "151671", "151672", "151673",
        "151674", "151675", "151676"]
n_clusters = [7] * 4 + [5] * 4 + [7] * 4
ARIs = []
refinement = True
for i in range(12):
  dataset = samples[i]
  file_fold = "/media/s5/gj/nodecode/data/DLPFC/{}".format(dataset)
  adata = sc.read_visium(file_fold, count_file='{}_filtered_feature_bc_matrix.h5'.format(dataset), load_images=True)
  adata.var_names_make_unique()
  # define model
  model = GraphST.GraphST(adata, device=device)
  # train model
  adata = model.train()
  # clustering
  # For DLPFC dataset, we use optional refinement step.
  clustering(adata, n_clusters[i], radius=50, method='mclust', refinement=refinement)
  # add ground_truth
  df_meta = pd.read_csv(file_fold + '/gt/{}_truth.txt'.format(dataset), sep='\t', header=None, index_col=0)
  adata.obs['ground_truth'] = df_meta.values
  # filter out NA nodes
  adata = adata[~pd.isnull(adata.obs['ground_truth'])]
  # calculate metric ARI
  ARI = metrics.adjusted_rand_score(adata.obs['domain'], adata.obs['ground_truth'])
  ARIs.append(round(ARI, 2))
print(ARIs)
print("mean:", round(np.mean(ARIs), 2))

My reproduction results:

[0.43, 0.5, 0.53, 0.47, 0.59, 0.43, 0.61, 0.64, 0.64, 0.62, 0.6, 0.43]
mean: 0.54