from fastai.basic_train import *
from fastai.gen_doc.nbdoc import *
from fastai.vision import *
from fastai.distributed import *

show_doc(Learner, title_level=2)

path = untar_data(URLs.MNIST_SAMPLE)
data = ImageDataBunch.from_folder(path)
learn = cnn_learner(data, models.resnet18, metrics=accuracy)

show_doc(Learner.lr_find)

learn.lr_find()

learn.recorder.plot()

show_doc(Learner.fit)

learn.fit(1)

show_doc(Learner.fit_one_cycle)

learn.fit_one_cycle(1)

show_doc(Learner.predict)

learn.data.train_ds[0]

data = learn.data.train_ds[0][0]

data

pred = learn.predict(data)
pred

learn.data.classes

probs = pred[2]

learn.data.valid_ds[0][0]

show_doc(Learner.get_preds)

learn.data.batch_size

preds = learn.get_preds()
preds

preds[0]

preds[1]

preds[1][0]

len(learn.data.valid_ds)

len(preds[0]), len(preds[1])

learn.get_preds(ds_type=DatasetType.Train)

learn.get_preds(with_loss=True)

show_doc(Learner.validate)

str(learn.metrics)

learn.validate()

learn.validate(learn.data.valid_dl)

learn.validate(learn.data.train_dl)

show_doc(Learner.show_results)

learn.show_results()

learn.show_results(ds_type=DatasetType.Train)

show_doc(Learner.pred_batch)

learn.data.batch_size

preds = learn.pred_batch()
len(preds)

preds[:10]

item = learn.data.train_ds[0][0]
item

batch = learn.data.one_item(item)
batch

learn.pred_batch(batch=batch)

show_doc(Learner.interpret, full_name='interpret')

jekyll_note('This function only works in the vision application.')

show_doc(Learner.summary)

show_doc(Learner.TTA, full_name = 'TTA')

show_doc(Learner.clip_grad)

show_doc(Learner.to_fp16)

show_doc(Learner.to_fp32)

show_doc(Learner.to_distributed, full_name='to_distributed')

show_doc(Learner.to_parallel, full_name='to_parallel')

# creates 3 layer groups
learn.split(lambda m: (m[0][6], m[1]))
# only randomly initialized head now trainable
learn.freeze()

learn.fit_one_cycle(1)

# all layers now trainable
learn.unfreeze()
# optionally, separate LR and WD for each group
learn.fit_one_cycle(1, max_lr=(1e-4, 1e-3, 1e-2), wd=(1e-4,1e-4,1e-1))

show_doc(Learner.lr_range)

learn.lr_range(slice(1e-5,1e-3)), learn.lr_range(slice(1e-3))

show_doc(Learner.unfreeze)

show_doc(Learner.freeze)

show_doc(Learner.freeze_to)

show_doc(Learner.split)

show_doc(Learner.save)

learn.save("trained_model")

learn.save("trained_model", return_path=True)

show_doc(Learner.load)

learn = learn.load("trained_model")

show_doc(Learner.export)

learn.export()

learn.export('trained_model.pkl')

path = learn.path
path

show_doc(load_learner)

learn = load_learner(path)

learn = load_learner(path, 'trained_model.pkl')

show_doc(Learner.purge)

show_doc(Learner.destroy)

show_doc(Learner.init)

show_doc(Learner.mixup)

show_doc(Learner.backward)

show_doc(Learner.create_opt)

show_doc(Learner.dl)

learn.dl()

learn.dl(DatasetType.Train)

show_doc(Recorder, title_level=2)

show_doc(Recorder.plot)

path = untar_data(URLs.MNIST_SAMPLE)
data = ImageDataBunch.from_folder(path)
learn = cnn_learner(data, models.resnet18, metrics=accuracy)

learn.lr_find()
learn.recorder.plot()

show_doc(Recorder.plot_losses)

learn.fit_one_cycle(5)
learn.recorder.plot_losses()

show_doc(Recorder.plot_lr)

learn.recorder.plot_lr()

learn.recorder.plot_lr(show_moms=True)

show_doc(Recorder.plot_metrics)

learn.recorder.plot_metrics()

show_doc(Recorder.on_backward_begin)

show_doc(Recorder.on_batch_begin)

show_doc(Recorder.on_epoch_end)

show_doc(Recorder.on_train_begin)

show_doc(Recorder.add_metric_names)

show_doc(Recorder.format_stats)

show_doc(fit)

show_doc(train_epoch)

show_doc(validate)

show_doc(get_preds)

show_doc(loss_batch)

show_doc(LearnerCallback, title_level=3)

show_doc(RecordOnCPU, title_level=3)

show_doc(Learner.tta_only)

show_doc(Learner.TTA)

show_doc(RecordOnCPU.on_batch_begin)