From c1356bffa3e2d1f87026024045d1fb9827622241 Mon Sep 17 00:00:00 2001
From: Yandi <yandirzm@gmail.com>
Date: Sun, 22 Jan 2023 18:41:45 +0100
Subject: [PATCH] [Normalizing the input]
---
create_submission.py | 1 -
dataloader.py | 1 +
main.py | 15 ++++++++++-----
train.py | 17 ++++++++++-------
4 files changed, 21 insertions(+), 13 deletions(-)
diff --git a/create_submission.py b/create_submission.py
index 126caf8..935934f 100644
--- a/create_submission.py
+++ b/create_submission.py
@@ -71,7 +71,6 @@ def create_submission(model, transform, device):
with torch.no_grad():
for X in tqdm.tqdm(test_loader):
X = X.to(device)
- print(X.shape)
#############################################
# This is where you inject your knowledge
# About your model
diff --git a/dataloader.py b/dataloader.py
index 0f131ea..9c9df4d 100644
--- a/dataloader.py
+++ b/dataloader.py
@@ -121,6 +121,7 @@ def get_dataloaders(
target_transform=valid_target_transform,
num_days=num_days,
)
+
# The sum of the two folds are not expected to be exactly of
# max_num_samples
logging.info(f" - The train fold has {len(train_dataset)} samples")
diff --git a/main.py b/main.py
index a759e7e..9b2efb8 100644
--- a/main.py
+++ b/main.py
@@ -55,9 +55,9 @@ if __name__ == "__main__":
#network = network.build_network(cfg, 18)
network = nn.Sequential(
- nn.Linear(14,8,False),
+ nn.Linear(14,35,True),
nn.ReLU(),
- nn.Linear(8, 35, True),
+ nn.Linear(35, 35, True),
nn.ReLU(),
nn.Linear(35,35,True),
nn.ReLU(),
@@ -67,18 +67,23 @@ if __name__ == "__main__":
nn.ReLU(),
nn.Linear(35,35,True),
nn.ReLU(),
+ nn.Linear(35,35, True),
+ nn.ReLU(),
nn.Linear(35,1, True),
nn.ReLU()
)
+ """
def init_xavier(module):
if type(module)==nn.Linear:
- nn.init.xavier_uniform_(module.weight)
+ nn.init.constant_(module.weight,1)
+ """
+
network = network.to(device)
"""
- for param in list(network.parameters()):
- param = 1
+ for layer in network:
+ layer.apply(init_xavier)
"""
f_loss = losses.RMSLELoss()
diff --git a/train.py b/train.py
index ef930e9..cb594f8 100644
--- a/train.py
+++ b/train.py
@@ -1,6 +1,7 @@
from tqdm import tqdm
import matplotlib.pyplot as plt
import numpy as np
+import torch
def train(model, loader, f_loss, optimizer, device):
"""
@@ -35,17 +36,19 @@ def train(model, loader, f_loss, optimizer, device):
optimizer.zero_grad()
loss.backward()
+
+ #torch.nn.utils.clip_grad_norm(model.parameters(), 50)
- #Y = list(model.parameters())[0].grad.cpu().tolist()
+ Y = list(model.parameters())[0].grad.cpu().tolist()
- #gradients.append(np.mean(Y))
- #tar.append(np.mean(outputs.cpu().tolist()))
- #out.append(np.mean(targets.cpu().tolist()))
+ gradients.append(np.mean(Y))
+ tar.append(np.mean(outputs.cpu().tolist()))
+ out.append(np.mean(targets.cpu().tolist()))
optimizer.step()
- #visualize_gradients(gradients)
- #visualize_gradients(tar)
- #visualize_gradients(out)
+ visualize_gradients(gradients)
+ visualize_gradients(tar)
+ visualize_gradients(out)
def visualize_gradients(gradients):
print(gradients)
--
GitLab