Skip to content

Commit 2fb0791

Browse files
bartzbeielsteinbartzbeielstein
committed
v0.0.60
1 parent 8dccae7 commit 2fb0791

4 files changed

Lines changed: 31 additions & 133 deletions

File tree

‎pyproject.toml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
77

88
[project]
99
name = "spotPython"
10-
version = "0.0.58"
10+
version = "0.0.60"
1111
authors = [
1212
{ name="T. Bartz-Beielstein", email="tbb@bartzundbartz.de" }
1313
]

‎src/spotPython/torch/netcifar10.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,9 @@
11
from torch import nn
22
import torch.nn.functional as F
3-
import spotPython.torch.netcorecv as netcorecv
3+
import spotPython.torch.netcore as netcore
44

55

6-
class Net_CIFAR10(netcorecv.Net_Core_CV):
6+
class Net_CIFAR10(netcore.Net_Core):
77
def __init__(self, l1, l2, lr, batch_size, epochs, k_folds):
88
super(Net_CIFAR10, self).__init__(lr=lr, batch_size=batch_size, epochs=epochs, k_folds=k_folds)
99
self.conv1 = nn.Conv2d(3, 6, 5)

src/spotPython/torch/netcorecv.py renamed to src/spotPython/torch/netcore.py

Lines changed: 26 additions & 128 deletions
Original file line numberDiff line numberDiff line change
@@ -7,134 +7,15 @@
77
from torch.utils.data import random_split
88

99

10-
class Net_Core_CV(nn.Module):
10+
class Net_Core(nn.Module):
1111
def __init__(self, lr, batch_size, epochs, k_folds):
12-
super(Net_Core_CV, self).__init__()
12+
super(Net_Core, self).__init__()
1313
self.lr = lr
1414
self.batch_size = batch_size
1515
self.epochs = epochs
1616
self.k_folds = k_folds
1717
self.results = {}
1818

19-
# def evaluate_cv_old(self, dataset, shuffle=False):
20-
# try:
21-
# device = getDevice()
22-
# self.to(device)
23-
# if torch.cuda.device_count() > 1:
24-
# self = nn.DataParallel(self)
25-
# criterion = nn.CrossEntropyLoss()
26-
# optimizer = optim.SGD(self.parameters(), lr=self.lr, momentum=0.9)
27-
# # TODO:
28-
# # if checkpoint_dir:
29-
# # model_state, optimizer_state = torch.load(os.path.join(checkpoint_dir, "checkpoint"))
30-
# # model.load_state_dict(model_state)
31-
# # optimizer.load_state_dict(optimizer_state)
32-
# # TODO:
33-
# # trainset, testset = load_data(data_dir)
34-
# # dataset = fun_control["train"]
35-
# kfold = KFold(n_splits=self.k_folds, shuffle=shuffle)
36-
37-
# # test_abs = int(len(dataset) * 0.6)
38-
# # train_subset, val_subset = random_split(dataset, [test_abs, len(dataset) - test_abs])
39-
# for fold, (train_ids, val_ids) in enumerate(kfold.split(dataset)):
40-
# print(f"Fold {fold}")
41-
# # Sample elements randomly from a given list of ids, no replacement.
42-
# train_subsampler = torch.utils.data.SubsetRandomSampler(train_ids)
43-
# val_subsampler = torch.utils.data.SubsetRandomSampler(val_ids)
44-
# # Define data loaders for training and testing data in this fold
45-
# trainloader = torch.utils.data.DataLoader(dataset,
46-
# batch_size=self.batch_size, sampler=train_subsampler)
47-
# valloader = torch.utils.data.DataLoader(dataset,
48-
# batch_size=self.batch_size, sampler=val_subsampler)
49-
# self.reset_weights()
50-
# # Define best_score, counter, and patience for early stopping:
51-
# best_score = None
52-
# counter = 0
53-
# patience = 10
54-
# # path = os.path.join(".", "checkpoint")
55-
# for epoch in range(self.epochs): # loop over the dataset multiple times
56-
# running_loss = 0.0
57-
# epoch_steps = 0
58-
# for i, data in enumerate(trainloader, 0):
59-
# # get the inputs; data is a list of [inputs, labels]
60-
# inputs, labels = data
61-
# inputs, labels = inputs.to(device), labels.to(device)
62-
63-
# # zero the parameter gradients
64-
# optimizer.zero_grad()
65-
66-
# # forward + backward + optimize
67-
# outputs = self(inputs)
68-
# loss = criterion(outputs, labels)
69-
# loss.backward()
70-
# optimizer.step()
71-
72-
# # print statistics
73-
# running_loss += loss.item()
74-
# epoch_steps += 1
75-
# if i % 2000 == 1999: # print every 2000 mini-batches
76-
# print("[%d, %5d] loss: %.3f" % (epoch + 1, i + 1, running_loss / epoch_steps))
77-
# running_loss = 0.0
78-
79-
# # Validation loss
80-
# val_loss = 0.0
81-
# val_steps = 0
82-
# total = 0
83-
# correct = 0
84-
# for i, data in enumerate(valloader, 0):
85-
# with torch.no_grad():
86-
# inputs, labels = data
87-
# inputs, labels = inputs.to(device), labels.to(device)
88-
89-
# outputs = self(inputs)
90-
# _, predicted = torch.max(outputs.data, 1)
91-
# total += labels.size(0)
92-
# correct += (predicted == labels).sum().item()
93-
# loss = criterion(outputs, labels)
94-
# val_loss += loss.cpu().numpy()
95-
# val_steps += 1
96-
# # Print accuracy
97-
# print("Accuracy for fold %d: %d %%" % (fold, 100.0 * correct / total))
98-
# print("--------------------------------")
99-
# self.results[fold] = 100.0 * (correct / total)
100-
# # early stopping:
101-
# # https://stackoverflow.com/questions/60200088/
102-
# how-to-make-early-stopping-in-image-classification-pytorch
103-
# if best_score is None:
104-
# best_score = val_loss
105-
# else:
106-
# # Check if val_loss improves or not.
107-
# if val_loss < best_score:
108-
# # val_loss improves, we update the latest best_score,
109-
# # and save the current model
110-
# best_score = val_loss
111-
# # TODO:
112-
# # torch.save({'state_dict':self.state_dict()}, path)
113-
# else:
114-
# # val_loss does not improve, we increase the counter,
115-
# # stop training if it exceeds the amount of patience
116-
# counter += 1
117-
# if counter >= patience:
118-
# break
119-
# # TODO:
120-
# # torch.save((self.state_dict(), optimizer.state_dict()), path)
121-
# # Print fold results
122-
# print(f"k-fold CV results for {self.k_folds} folds")
123-
# print("--------------------------------")
124-
# sum = 0.0
125-
# for key, value in self.results.items():
126-
# print(f"Fold {key}: {value} %")
127-
# sum += value
128-
# avg = sum / len(self.results.items())
129-
# print(f"Average: {avg} %")
130-
# df_eval = avg
131-
# df_preds = np.nan
132-
# except Exception as err:
133-
# print(f"Error in Net_Core. Call to evaluate() failed. {err=}, {type(err)=}")
134-
# df_eval = np.nan
135-
# df_preds = np.nan
136-
# return df_eval, df_preds
137-
13819
def reset_weights(self):
13920
for layer in self.children():
14021
if hasattr(layer, "reset_parameters"):
@@ -195,7 +76,7 @@ def evaluate_cv(self, dataset, shuffle=False):
19576
df_eval = sum(self.results.values()) / len(self.results.values())
19677
df_preds = np.nan
19778
except Exception as err:
198-
print(f"Error in Net_Core_CV. Call to evaluate_cv() failed. {err=}, {type(err)=}")
79+
print(f"Error in Net_Core. Call to evaluate_cv() failed. {err=}, {type(err)=}")
19980
df_eval = np.nan
20081
df_preds = np.nan
20182
return df_eval, df_preds
@@ -213,16 +94,30 @@ def evaluate_hold_out(self, dataset, shuffle, test_dataset=None):
21394
else:
21495
trainloader, valloader = self.create_train_test_data_loaders(dataset, shuffle, test_dataset)
21596
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.1)
97+
# Early stopping parameters
98+
patience = 5
99+
best_val_loss = float("inf")
100+
counter = 0
216101
for epoch in range(epochs):
217102
self.train_hold_out(trainloader, criterion, optimizer, device=device, epoch=epoch)
218103
scheduler.step()
219-
val_accuracy, val_loss = self.validate_hold_out(valloader=valloader, criterion=criterion, device=device)
220-
df_eval = val_loss
104+
# Early stopping check
105+
val_accuracy, val_loss = self.validate_hold_out(valloader=valloader, criterion=criterion, device=device)
106+
if val_loss < best_val_loss:
107+
best_val_loss = val_loss
108+
counter = 0
109+
else:
110+
counter += 1
111+
if counter >= patience:
112+
print(f"Early stopping at epoch {epoch}")
113+
break
114+
df_eval = best_val_loss
221115
df_preds = np.nan
222116
except Exception as err:
223-
print(f"Error in Net_Core_CV. Call to evaluate_hold_out() failed. {err=}, {type(err)=}")
117+
print(f"Error in Net_Core. Call to evaluate_hold_out() failed. {err=}, {type(err)=}")
224118
df_eval = np.nan
225119
df_preds = np.nan
120+
print(f"Returned to Spot: Best validation loss: {df_eval}")
226121
return df_eval, df_preds
227122

228123
def create_train_val_data_loaders(self, dataset, shuffle):
@@ -260,8 +155,11 @@ def train_hold_out(self, trainloader, criterion, optimizer, device, epoch):
260155
# print statistics
261156
running_loss += loss.item()
262157
epoch_steps += 1
263-
if i % 2000 == 1999: # print every 2000 mini-batches
264-
print("[%d, %5d] loss: %.3f" % (epoch + 1, i + 1, running_loss / epoch_steps))
158+
if i % 1000 == 999: # print every 1000 mini-batches
159+
print(
160+
"Epoch: %d, Batch: %5d. Batch Size: %d. Training Loss: %.3f"
161+
% (epoch + 1, i + 1, int(self.batch_size), running_loss / epoch_steps)
162+
)
265163
running_loss = 0.0
266164

267165
def validate_hold_out(self, valloader, criterion, device):
@@ -284,6 +182,6 @@ def validate_hold_out(self, valloader, criterion, device):
284182
val_steps += 1
285183
accuracy = correct / total
286184
loss = val_loss / val_steps
287-
print(f"Accuracy on hold-out set: {accuracy}")
288185
print(f"Loss on hold-out set: {loss}")
186+
print(f"Accuracy on hold-out set: {accuracy}")
289187
return accuracy, loss

‎src/spotPython/torch/netcvfashionMNIST.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,8 @@
11
from torch import nn
2-
import spotPython.torch.netcorecv as netcorecv
2+
import spotPython.torch.netcore as netcore
33

44

5-
class Net_CV_fashionMNIST(netcorecv.Net_Core_CV):
5+
class Net_CV_fashionMNIST(netcore.Net_Core):
66
def __init__(self, l1, l2, lr, batch_size, epochs, k_folds):
77
super(Net_CV_fashionMNIST, self).__init__(lr=lr, batch_size=batch_size, epochs=epochs, k_folds=k_folds)
88
self.flatten = nn.Flatten()

0 commit comments

Comments
 (0)