The correspondence between each row of the latent_codes_state_dict['weight'] matrix and id and expression

[self-hacker]

How to check which id and expression each row of the latent_codes_state_dict['weight'] matrix in a checkpoint of a certain epoch corresponds to?
I tried adding some code to the get_loader function in face_dataset.py to try to match them to each row by looking at the id and expr loading order in the batch, but I found that the loading order is different for each run, and there are repeated idx being loaded. Is the random seed somewhere not fixed?
What should I do to see the correspondence between the latent_codes_state_dict['weight'] matrix and id and expression
def get_loader(self, shuffle=True):
random.seed(0)
torch.manual_seed(0)
torch.cuda.manual_seed(0)
np.random.seed(0)

    print("Sampler order:")
    dataloader=torch.utils.data.DataLoader(
        self, batch_size=self.batch_size, num_workers=1, shuffle=shuffle,
        worker_init_fn=self.worker_init_fn,
        pin_memory=True)
    
    import pandas as pd
    id_expression_map = []
    for batch in dataloader:
        batch_ids = batch['iden']
        batch_expressions = batch['expr']
        # 记录每个样本的顺序
        id_expression_map.extend(zip(batch_ids, batch_expressions))

    # 将数据转换为 DataFrame，以保证行顺序一致
    id_expr_df = pd.DataFrame(id_expression_map, columns=['id', 'expression'])
    print(id_expr_df)

    return torch.utils.data.DataLoader(
        self, batch_size=self.batch_size, num_workers=1, shuffle=shuffle,
        worker_init_fn=self.worker_init_fn,
        pin_memory=True)