sub = pd.read_csv('data/sample_submission.csv')
sub['cell_line'] = preds
sub.to_csv('save_6.csv', index=False)
trainers.report_result()
from sklearn import metrics
import matplotlib.pyplot as plt
import seaborn as sns
colors = ["windows blue", "amber", "greyish", "faded green", "dusty purple"]
sns.palplot(sns.xkcd_palette(colors))
_, _, (y_true, y_pred) = trainers.evaluate(val_loader)
cm = metrics.confusion_matrix(y_true, y_pred, normalize='true')
idx = sorted(np.unique(y_true))
confusion_matrix = pd.DataFrame(cm,
index=process.enc.classes_,
columns=process.enc.classes_)
plt.figure(figsize=(7, 7))
sns.heatmap(confusion_matrix,
annot=True,
fmt=".2f",
square=True,
cbar=False,
cmap=sns.color_palette("Blues"))
plt.show()
plot_cm(val_loader)
print(
"<%s> Epoch: %03d/%03d, Iteration: %03d/%03d, Loss: %.8f, Acc: %.3f"
% (get_local_time(), epoch + 1, config['n_epochs'], it + 1,
len(train_loader), loss, acc))
iterations = epoch * len(train_loader) + it + 1
write_loss(iterations, trainer, train_writer)
if (epoch + 1) % config['test_iter'] == 0:
t_bar = tqdm.tqdm(test_loader)
t_bar.set_description('Epoch: {} - Testing'.format(epoch + 1))
losses = []
accuracy_list = []
for (images, labels) in t_bar:
images = images.cuda()
labels = labels.cuda()
loss, accuracy = trainer.evaluate(images, labels)
losses.append(loss)
accuracy_list.append(accuracy)
mean_loss = np.mean(losses)
accuracy = np.mean(accuracy_list)
print('\n<{}> Test result: loss: {}, accuracy: {}'.format(
get_local_time(), mean_loss, accuracy))
if opts.fine_tune:
trainer.save(checkpoint_directory,
epoch,
acc=accuracy,
min_loss=mean_loss,
post_fix='-{}'.format(epoch + 1 - epochs))
print('\n<{}> Saving the ft model, the prediction accuracy: {}'.
format(get_local_time(), accuracy))
if accuracy > max_acc: