From 8e0e63cbd16c3797abd5f3a4f9f5afd0c77d188a Mon Sep 17 00:00:00 2001
From: awehenkel <antoine.wehenkel@gmail.com>
Date: Sat, 18 Apr 2020 16:16:12 +0200
Subject: [PATCH] Correct mistake with parameters instead of buffers.

---
 models/UMNN/UMNNMAF.py     | 7 ++-----
 models/UMNN/UMNNMAFFlow.py | 4 ++--
 models/UMNN/__init__.py    | 4 +++-
 3 files changed, 7 insertions(+), 8 deletions(-)

diff --git a/models/UMNN/UMNNMAF.py b/models/UMNN/UMNNMAF.py
index a3bff49..bef11ce 100644
--- a/models/UMNN/UMNNMAF.py
+++ b/models/UMNN/UMNNMAF.py
@@ -41,19 +41,16 @@ def __init__(self, net, input_size, nb_steps=100, device="cpu", solver="CC"):
         self.cc_weights = None
         self.steps = None
         self.solver = solver
-        self.pi = nn.Parameter(torch.tensor(math.pi).to(self.device), requires_grad=False)
+        self.register_buffer("pi", torch.tensor(math.pi))
 
         # Scaling could be changed to be an autoregressive network output
         self.scaling = nn.Parameter(torch.zeros(input_size, device=self.device), requires_grad=False)
 
     def to(self, device):
         self.device = device
-        self.net.to(device)
-        self.pi = nn.Parameter(torch.tensor(math.pi).to(self.device), requires_grad=False)
-        self.scaling = self.scaling.to(self.device)
+        super().to(device)
         return self
 
-
     def forward(self, x, method=None, x0=None, context=None):
         x0 = x0.to(x.device) if x0 is not None else torch.zeros(x.shape).to(x.device)
         xT = x
diff --git a/models/UMNN/UMNNMAFFlow.py b/models/UMNN/UMNNMAFFlow.py
index e78b579..4ff1952 100644
--- a/models/UMNN/UMNNMAFFlow.py
+++ b/models/UMNN/UMNNMAFFlow.py
@@ -53,7 +53,7 @@ def __init__(self, nb_flow=1, nb_in=1, hidden_derivative=[50, 50, 50, 50], hidde
         """
         super().__init__()
         self.device = device
-        self.pi = nn.Parameter(torch.tensor(math.pi, device=self.device), requires_grad=False)
+        self.register_buffer("pi", torch.tensor(math.pi))
         self.nets = ListModule(self, "Flow")
         for i in range(nb_flow):
             auto_net = EmbeddingNetwork(nb_in, hidden_embedding, hidden_derivative, embedding_s, act_func=act_func,
@@ -66,7 +66,7 @@ def to(self, device):
         for net in self.nets:
             net.to(device)
         self.device = device
-        self.pi = self.pi.to(device)
+        super().to(device)
         return self
 
     def forward(self, x, context=None):
diff --git a/models/UMNN/__init__.py b/models/UMNN/__init__.py
index b73bcfc..38d3c1a 100644
--- a/models/UMNN/__init__.py
+++ b/models/UMNN/__init__.py
@@ -1,4 +1,6 @@
 from .UMNNMAFFlow import UMNNMAFFlow
 from .MonotonicNN import MonotonicNN, IntegrandNN
 from .UMNNMAF import IntegrandNetwork, UMNNMAF
-from .made import MADE
\ No newline at end of file
+from .made import MADE
+from .NeuralIntegral import NeuralIntegral
+from .ParallelNeuralIntegral import ParallelNeuralIntegral
\ No newline at end of file