def plot_train_val(patterns, fontsize=15):
import matplotlib.pyplot as plt
import seaborn as sns
import matplotlib.ticker as ticker
dummy_trainer = Train(None, None, None, None)
for pattern in patterns:
for cpt_fn in glob.glob(pattern):
cpt = torch.load(cpt_fn)
name = cpt_fn.split('.')[0].split('/')[-1]
l = cpt['train_losses']
dummy_trainer.all_losses = l
plt.plot(dummy_trainer.smooth_loss(), label=name)
plt.legend(loc='center right', bbox_to_anchor=(1.5, 0.5))
plt.title('training loss (RMSE)', fontsize=fontsize)
plt.grid()
plt.show()
for pattern in patterns:
for cpt_fn in glob.glob(pattern):
cpt = torch.load(cpt_fn)
name = cpt_fn.split('.')[0].split('/')[-1]
l = cpt['val_losses']
dummy_trainer.val_losses = l
plt.plot(dummy_trainer.smooth_valloss(), label=name)
plt.legend(loc='center right', bbox_to_anchor=(1.5, 0.5))
plt.title('validation loss (RMSE)', fontsize=fontsize)
plt.grid()
plt.show()
def get_train_val_curves(pattern):
dummy_trainer = Train(None, None, None, None)
tr_curves = []
val_curves = []
name = ""
for cpt_fn in glob.glob(pattern):
cpt = torch.load(cpt_fn)
name = cpt_fn.split('.')[0].split('/')[-1]
dummy_trainer.all_losses = cpt['train_losses']
dummy_trainer.val_losses = cpt['val_losses']
tr_curves.append(dummy_trainer.smooth_loss())
val_curves.append(dummy_trainer.smooth_valloss())
return tr_curves, val_curves, name