o
oi! @ s d dl Z d dlmZmZmZmZ d dlZd dlmZm Z d dl
m Z zd dl m
Z
W n ey5 dZ
Y nw d dlmZmZ d dlmZ dd lmZmZmZ g d
ZG d d d ZdS )
N)AnyCallableDictOptional) Optimizer lr_scheduler)
DataLoader) Accelerator)ModuleTensor) AverageMeter )
Configuration StatsTracker TrainerState)
preprocess
augmentationsevaluatefit fit_epochon_epoch_starton_before_modelon_after_model
on_checkpoint on_epoch_endc @ s e Zd ZdZ d4dedee dee dee dede j
d e d
ee e
ed f d dfd
dZed ejfddZded dfddZded dfddZd5ddZe d e e
ef fddZded ed dfd d Zd e e
ef d e e
ef fd d!Z d e e
ef d e e
ef fd"d#Z ded e e
e f fd$d%Z ded efd&d'Z d e e
ef d e e
ef fd(d)Z!ded*e e
ef d efd+d,Z"d-ed*e e
ef d dfd.d/Z#ded e e
ef d dfd0d1Z$ded e e
ef d dfd2d3Z%dS )6Trainera Base class to train the different models in kornia.
.. warning::
The API is experimental and subject to be modified based on the needs of kornia models.
Args:
model: the nn.Module to be optimized.
train_dataloader: the data loader used in the training loop.
valid_dataloader: the data loader used in the validation loop.
criterion: the nn.Module with the function that computes the loss.
optimizer: the torch optimizer object to be used during the optimization.
scheduler: the torch scheduler object with defiing the scheduling strategy.
accelerator: the Accelerator object to distribute the training.
config: a TrainerConfiguration structure containing the experiment hyper parameters.
callbacks: a dictionary containing the pointers to the functions to overrides. The
main supported hooks are ``evaluate``, ``preprocess``, ``augmentations`` and ``fit``.
.. important::
The API heavily relies on `accelerate `_.
In order to use it, you must: ``pip install kornia[x]``
.. seealso::
Learn how to use the API in our documentation
`here `_.
Nmodeltrain_dataloadervalid_dataloader criterion optimizer schedulerconfig callbacks).Nreturnc C s t d u rtdt | _| j|| _| j|| _| j|| _|d u r'd n|| j| _ | j|| _
|| _ || _ |d u rAi }|
D ]\} }
| tvrUtd| dtt| |
qE|j| _tj| _td| _d S )Nz } s zTrainer.backwardepochc C s | j t }t| jD ]v\}}|d |d d}| j | |}| |}| |}|
| j |}| || | ||d }|
| | j || t|d |d dkr| jd|d d| j d |d dt| j d
|jd d |jd q
d S )
Nr r
inputtargetrB rA 2 zTrain: /z
Sample: z Loss: z.3f )r r& r enumerater r zero_gradr r r on_modelr compute_lossr> stepupdateitemlenr5 infor0 valavg)r6 r? losses sample_idsampleoutputr= r9 r9 r: r s8
zTrainer.fit_epochc C sj t | jD ]-}tj| _| | tj| _| }| | j || |
| jtj kr- d S | j
qd S r< )ranger0 r TRAININGr2 r VALIDATEr r r r TERMINATEr! rJ )r6 r? valid_statsr9 r9 r: r s
z Trainer.fitc C s | j t }t| jD ]d\}}|d |d d}| |}| |}| | j |}| || t |d }| j
d urM| ||d }| d|
| || ||d | |d dkrq| jd| d t | j d
| q
| S ) Nr r
r@ rA rB rQ
zTest: rD rE )r evalr rF r r r rH r rM r rI rK rL update_from_dictcompute_metricsr5 rN as_dict)r6 statsrR rS out
batch_sizeval_lossr9 r9 r: r s"
$zTrainer.evaluateargskwargsc O d S r< r9 r6 rc rd r9 r9 r: r zTrainer.on_epoch_startxc C |S r< r9 r6 rh r9 r9 r: r zTrainer.preprocessc C ri r< r9 rj r9 r9 r: r rk zTrainer.augmentationsc G s i S )z&Compute metrics during the evaluation.r9 r6 rc r9 r9 r: r] s zTrainer.compute_metricsc G s | j d u r td| j | S )Nz `criterion` should not be None.)r RuntimeErrorrl r9 r9 r: rI s
zTrainer.compute_lossc C ri r< r9 rj r9 r9 r: r rk zTrainer.on_before_modelrS c C s ||d S )NrA r9 )r6 r rS r9 r9 r: rH s zTrainer.on_modelrT c C re r< r9 )r6 rT rS r9 r9 r: r rg zTrainer.on_after_modelc O re r< r9 rf r9 r9 r: r rg zTrainer.on_checkpointc O re r< r9 rf r9 r9 r: r rg zTrainer.on_epoch_endr< )r$ N)&__name__
__module__ __qualname____doc__r
r r r r r _LRSchedulerr r strr r; propertytorchr+ r r> intr r no_gradr r r r r floatr] rI r rH r r r r9 r9 r9 r: r 5 sN $
(
""" "r )r3 typingr r r r ru torch.optimr r torch.utils.datar
accelerater ImportError kornia.corer
r kornia.metricsr utilsr r r r- r r9 r9 r9 r: s