From 40c427306bde690a39f22f75ad2db52927eff5c9 Mon Sep 17 00:00:00 2001
From: Joerg Martin <joerg.martin@ptb.de>
Date: Fri, 3 Dec 2021 14:43:20 +0100
Subject: [PATCH] Fixed bug in prediction method of FNNEIV
---
EIVPackage/EIVArchitectures/Networks.py | 4 ++--
Experiments/evaluate_tabular.py | 6 ++++--
2 files changed, 6 insertions(+), 4 deletions(-)
diff --git a/EIVPackage/EIVArchitectures/Networks.py b/EIVPackage/EIVArchitectures/Networks.py
index 3a26b02..09d75b8 100644
--- a/EIVPackage/EIVArchitectures/Networks.py
+++ b/EIVPackage/EIVArchitectures/Networks.py
@@ -206,7 +206,7 @@ class FNNEIV(nn.Module):
pred, sigma = reshape_to_chunks(pred, sigma,
number_of_draws=parameter_sample_size * number_of_draws[1])
pred_collection.append(pred)
- sigma_collection.append(pred)
+ sigma_collection.append(sigma)
remaining_draws -= parameter_sample_size
pred = torch.cat(pred_collection, dim=1)
sigma = torch.cat(sigma_collection, dim=1)
@@ -279,7 +279,7 @@ class FNNEIV(nn.Module):
# average over parameter values
predictive_log_density_values = \
torch.logsumexp(input=exp_arg, dim=1)\
- - torch.log(torch.tensor(number_of_draws))
+ - torch.log(torch.prod(torch.tensor(number_of_draws)))
if average_batch_dimension:
return torch.mean(predictive_log_density_values, dim=0)
else:
diff --git a/Experiments/evaluate_tabular.py b/Experiments/evaluate_tabular.py
index 90b972f..e13f9d5 100644
--- a/Experiments/evaluate_tabular.py
+++ b/Experiments/evaluate_tabular.py
@@ -25,7 +25,7 @@ input_dim = train_data[0][0].numel()
output_dim = train_data[0][1].numel()
def collect_metrics(x,y, seed=0,
- noneiv_number_of_draws=100, eiv_number_of_draws=100,
+ noneiv_number_of_draws=100, eiv_number_of_draws=[100,1],
decouple_dimensions=False, device=torch.device('cuda:1')):
"""
:param x: A torch.tensor, taken as input
@@ -128,7 +128,8 @@ def collect_metrics(x,y, seed=0,
# NLL
training_state = net.training
net.train()
- eiv_logdens = net.predictive_logdensity(x, y, number_of_draws=100,
+ eiv_logdens = net.predictive_logdensity(x, y,
+ number_of_draws=eiv_number_of_draws,
decouple_dimensions=decouple_dimensions,
scale_labels=\
train_data.dataset.std_labels.view((-1,)).to(device)\
@@ -162,6 +163,7 @@ for seed in tqdm(seed_list):
# TODO: Despite statistics, the fluctuations seem to be large
+# TODO: fix sqrt scaling, missing factor
print('Non-EiV')
print(f'RMSE {np.mean(noneiv_rmse_collection):.3f}'\
f'({np.std(noneiv_rmse_collection)/np.sqrt(num_test_epochs):.3f})')
--
GitLab