Learner

Module containing helper functions and classes around the Learner

---

source

PublishEvents

 PublishEvents (name)

Initialize self. See help(type(self)) for accurate signature.

---

source

CancelBatchException

Common base class for all non-exit exceptions.

---

source

CancelEpochException

Common base class for all non-exit exceptions.

---

source

CancelFitException

Common base class for all non-exit exceptions.

---

source

Learner

 Learner (model, dls, loss_fn, optim_class, lr, subs)

Initialize self. See help(type(self)) for accurate signature.

---

source

Subscriber

 Subscriber ()

Initialize self. See help(type(self)) for accurate signature.

---

source

MetricsS

 MetricsS (**metrics)

Initialize self. See help(type(self)) for accurate signature.

---

source

DeviceS

 DeviceS (device)

Initialize self. See help(type(self)) for accurate signature.

---

source

LRFindS

 LRFindS (mult=1.25)

Initialize self. See help(type(self)) for accurate signature.

---

source

MomentumLearner

 MomentumLearner (model, dls, loss_fn, optim_class, lr, subs, mom=0.85)

Initialize self. See help(type(self)) for accurate signature.

---

source

ProgressS

 ProgressS (plot=False)

Initialize self. See help(type(self)) for accurate signature.

Example usage:

from datasets import load_dataset,load_dataset_builder

from nntrain.dataloaders import DataLoaders, hf_ds_collate_fn

name = "fashion_mnist"
ds_builder = load_dataset_builder(name)
hf_dd = load_dataset(name)

bs = 1024
n_in = 28*28
n_h = 50
n_out = 10
lr = 0.01

dls = DataLoaders.from_hf_dd(hf_dd, batch_size=bs)

Reusing dataset fashion_mnist (/root/.cache/huggingface/datasets/fashion_mnist/fashion_mnist/1.0.0/8d6c32399aa01613d96e2cbc9b13638f359ef62bb33612b077b4c247f6ef99c1)

def get_model():
    layers = [nn.Linear(n_in, n_h), nn.ReLU(), nn.Linear(n_h, n_out)]
    return nn.Sequential(*layers)

metrics = MetricsS(accuracy=tem.MulticlassAccuracy())
progress = ProgressS(True)
device = DeviceS(device)

l = MomentumLearner(get_model(), dls, F.cross_entropy, torch.optim.SGD, lr, [metrics, progress, device])
l.fit(5)

epoch	mode	loss	accuracy
0	train	1.763	0.458
0	eval	1.151	0.647
1	train	0.949	0.669
1	eval	0.846	0.685
2	train	0.777	0.719
2	eval	0.748	0.725
3	train	0.697	0.757
3	eval	0.683	0.762
4	train	0.643	0.780
4	eval	0.641	0.778