bird0401 commented 2 years ago

category classification model

entity classification model

entity linking only using imagenet and identify by its name using word similarity method
the number of images tend to be low and I should data augmentation on base model
- use imagenet and self-supervised learning because of small data in wikimedia commons

bird0401 commented 1 year ago

use Kaggle solution to similar problem
use simple and generally usable model
efficient-net
not use batch same normalization component repeatedly because it has parameters and it is part of learned target
try both fine tuing and transfer learning and compare them
use SGD because pre-trained model already has almost completed parameters, and it should not be changed

bird0401 commented 1 year ago

[ ] elastic distortion
[ ] mixup
normalization
[x] regularization
[x] unbalanced data
- [x] upsampling few images
[ ] label smoothing -> few improvement
others
[ ] delete zeros from input data due not to cancel out forward values
[x] use torchvision to download
[x] how to store image and label pairs
[x] split images to train, val, and test using csv from one directory
[x] use different transform to train and val
[x] delete classes that have less than 3 images
[x] define customed collate_fn function to handle corrupted images
[x] create dataframe containing file paths and labels referreing img directory
[x] create utility fuctions
[x] use imagefolder
- [x] when using imagefolder, augument should be transforms module , not albumentation
- [x] we should create wrapper function
- [x] we cant use imagefolder because we will do k_fold cross validation and split into a train and valid data using df and the procedure makes us take complex steps where dataset class should be used
[ ] delete some images
- [ ] delete image where there is no entity
- [ ] distinct duplicate images
- [ ] use auto deletion app
[x] to learn faster
- [x] set num_workers=2 instead of 0 in dataloader
- [x] 2 is the enough number
- [x] pin_memory = True
- [x] torch.backends.cudnn.benchmark = False not True
- [x] slower but keep reproductivity
[x] Stratified KFold to deal with imbalanced data

[x] grid search, optuna
- [x] params in CONFIG
[x] parallel computing
[x] use GPU effectively
[x] store model periodically
[x] show logs
[x] create early stopping module
- pytorch has not module of early stopping, but pytorch lightning has
[x] use lr_scheduler.ReduceLROnPlateau
[x] fine tuning using pre-trained model learned imagenet
[x] change data augumentation method
[ ] Stochastic Weight Averaging
[ ] use focal loss instead of cross entropy loss to deal with imbalance data
[x] transfer learning or fine tuning -> dont use transfer learning because it cant achieve high f1
- [x] requires_grad only in last layer

[ ] try some patterns of random seed when using not convex optimization
[x] use wandb
[x] use hydra
- [x] it enables config file to be written structurally
- [x] we can override values on command line
[x] create faster prototype and debug mode
- [x] small number of dataset
- [x] small number of classes
- [x] large batch_size
- [x] small epoch
- [x] half-precision training
- [x] as fast as twice or more, also as low memory as half
[ ] depth wise convs, instead of regular convs
[x] use learning rate scheduler
[x] use timm to import model
[x] use f1-macro because of imbalance data
[ ] show images and predicted labels
[ ] use vertex pipelines instead of kubeflow, mlflow, and airflow because it can execute on full-managed
- [ ] also instances will stand only when pipelines are executed
  Serving
[ ] pre-processing pipeline

bird0401 commented 1 year ago