def estimate_weights(fold):
holdout = folds == fold
train_datasets = []
for dset, has_fut in zip(datasets, has_future):
d = loglin.sub_dataset(dset, ~holdout & has_fut)
train_datasets.append(d)
weights = learn.learn_time_regularized_weights(train_datasets, censor_times, time_penalty=time_penalty, l1_penalty=l1_penalty)
return weights
for dset, has_fut in zip(datasets, has_future):
d = loglin.sub_dataset(dset, ~holdout & has_fut)
train_datasets.append(d)
weights = learn.learn_time_regularized_weights(train_datasets, censor_times, time_penalty=time_penalty, l1_penalty=l1_penalty)
return weights
pool = Pool(3)
all_weights = pool.map(estimate_weights, [k + 1 for k in range(10)])
for i, weights in enumerate(all_weights):
fold = i + 1
holdout = folds == fold
for j, w in enumerate(weights):
test_data = loglin.sub_dataset(datasets[j], holdout)
_, marg, _ = loglin.inference(w, test_data)
posteriors[j][holdout, :] = marg[0]
# for fold in sorted(set(folds)):
# logging.info('Staring fold {}'.format(fold))
# holdout = folds == fold
# train_datasets = []
# for dset, has_fut in zip(datasets, has_future):
# d = loglin.sub_dataset(dset, ~holdout & has_fut)
# train_datasets.append(d)
# all_weights = learn.learn_time_regularized_weights(train_datasets, censor_times, time_penalty=1e-3, l1_penalty=1e-3)
def estimate_weights(fold):
holdout = folds == fold
train_data = loglin.sub_dataset(dataset, ~holdout & has_future)
weights = learn.learn_weights(train_data, 1e-3)
