ARTICLE AD BOX
I have done 5 fold cross validation to create 5 fold: three for training data, one for validation and one for test data. I want to perform Kaplan-Meier method in this data.
I have computed 10 unique time points from the dataset:
def compute_eval_time(args): local_unique_times = [] for client_id in range(args.n_parties): train_df, val_df, _, _ = get_local_dataset(args.dataset, args.partition, client_id, args.run_id) train_time, _ = train_df["time"], train_df["event"] val_time, _ = val_df["time"], val_df["event"] local_unique_times.append(np.unique(np.concatenate([train_time, val_time]))) glob_unique_time=np.unique(np.concatenate(local_unique_times)) pseudo_evaltime=np.percentile(glob_unique_time, np.arange(args.pseudo_perc_interval, 99, args.pseudo_perc_interval)) return pseudo_evaltimeNow, I want to train the Kaplan-Meier method using my train and validation data and then want to evaluate the trained KM method using my test data. For this, I am using Kaplan-Meier Fitter from lifelines Python package.
def train_net_all(net_id, train_time_all, train_status_all, val_time_all, val_status_all, test_time_all, test_status_all, evaltime, args, device): logger.info('Training network %s' % str(net_id)) print('Training network %s' % str(net_id)) train_val_time = np.concatenate((train_time_all, val_time_all)) train_val_status = np.concatenate((train_status_all, val_status_all)) kmf = KaplanMeierFitter(label="waltons_data") kmf.fit(train_val_time, train_val_status) test_survival_probabilities = kmf.survival_function_at_times(test_time_all) ev1 = EvalSurv(test_survival_probabilities, test_time_all, test_status_all == 1, censor_surv='km') cindex = 0.5 # Integrated Brier score try: brier = float(ev1.integrated_brier_score(evaltime)) except Exception as e: print(f"[Brier ERROR] {type(e).__name__}: {e}") brier = 0.25 return cindex, brierWhat do you think? Is there any logical bug in my code? Please guide me. Any suggestions or feedback will be highly appreciated.
