def register(): registry.Criterion(RingLoss) registry.Callback(InferBestCallback) registry.Scheduler(OneCycleCosineAnnealLR) # classification try: from mlcomp.contrib.model import Pretrained registry.Model(Pretrained) except Exception: pass # segmentation registry.Model(Unet) registry.Model(ResnetLinknet) registry.Model(MobileUnet) registry.Model(ResnetUnet) registry.Model(ResnetFPNUnet) registry.Model(ResnetPSPnet) registry.Model(FPNUnet) registry.Model(Linknet) registry.Model(PSPnet) registry.Model(ResNetLinknet) try: from mlcomp.contrib.model.segmentation_model_pytorch import \ SegmentationModelPytorch registry.Model(SegmentationModelPytorch) except Exception: pass
def register(): registry.Model(Pretrained) registry.Criterion(RingLoss) registry.Callback(InferBestCallback) registry.Scheduler(OneCycleCosineAnnealLR) # segmentation registry.Model(Unet) registry.Model(ResnetLinknet) registry.Model(MobileUnet) registry.Model(ResnetUnet) registry.Model(ResnetFPNUnet) registry.Model(ResnetPSPnet) registry.Model(FPNUnet) registry.Model(Linknet) registry.Model(PSPnet)
# Register models registry.Model(finetune_vggresnet) registry.Model(DeepBranchResnet) registry.Model(DeepBranchAttResnet) registry.Model(DSInceptionV3) registry.Model(DSResnet) registry.Model(SSEResnet) registry.Model(DSResnetSA) registry.Model(DSResnetAttbranch) registry.Model(GAIN) registry.Model(GCAM) registry.Model(GAINMask) registry.Model(TemporalLSTM) # Register callbacks registry.Callback(LabelSmoothCriterionCallback) registry.Callback(SmoothMixupCallback) registry.Callback(DSAccuracyCallback) registry.Callback(DSCriterionCallback) registry.Callback(GAINCriterionCallback) registry.Callback(GAINSaveHeatmapCallback) registry.Callback(GCAMSaveHeatmapCallback) registry.Callback(GAINMaskCriterionCallback) # Register criterions registry.Criterion(LabelSmoothingCrossEntropy) # Register optimizers registry.Optimizer(AdamW) registry.Optimizer(Nadam)
GANLoss, GeneratorOptimizerCallback, IdenticalGANLoss, PrepareDiscriminatorPhase, PrepareGeneratorPhase, ) from src.callbacks.visualization import LogImageCallback from src.experiments.train.train_experiment import Experiment from src.modules.discriminator import NLayerDiscriminator, PixelDiscriminator from src.modules.generator import Generator from src.modules.loss import LSGanLoss from src.runner import CycleGANRunner as Runner registry.Model(Generator) registry.Model(PixelDiscriminator) registry.Model(NLayerDiscriminator) registry.Criterion(LSGanLoss) registry.Callback(CycleGANLoss) registry.Callback(GANLoss) registry.Callback(IdenticalGANLoss) registry.Callback(PrepareGeneratorPhase) registry.Callback(GeneratorOptimizerCallback) registry.Callback(PrepareGeneratorPhase) registry.Callback(PrepareDiscriminatorPhase) registry.Callback(DiscriminatorLoss) registry.Callback(DiscriminatorOptimizerCallback) registry.Callback(LogImageCallback)
registry.Model(SENetGrouplevel) registry.Model(EfficientNet) registry.Model(SENetTIMM) registry.Model(InceptionV3TIMM) registry.Model(GluonResnetTIMM) registry.Model(DSInceptionV3) registry.Model(DSSENet) registry.Model(DSResnet) registry.Model(ResNet50CutMix) registry.Model(Fishnet) registry.Model(SENetCellType) registry.Model(SENetCellMultipleDropout) registry.Model(MixNet) # Register callbacks registry.Callback(LabelSmoothCriterionCallback) registry.Callback(SmoothMixupCallback) registry.Callback(DSAccuracyCallback) registry.Callback(DSCriterionCallback) registry.Callback(SlackLogger) registry.Callback(TwoHeadsCriterionCallback) registry.Callback(DSMixupCallback) # Register criterions registry.Criterion(LabelSmoothingCrossEntropy) # Register optimizers registry.Optimizer(AdamW) registry.Optimizer(Nadam) registry.Optimizer(RAdam)
import experiments as exp if os.environ.get("USE_WANDB", "0") == "1": from catalyst.dl import SupervisedWandbRunner as Runner elif os.environ.get("USE_NEPTUNE", "0") == "1": from catalyst.dl import SupervisedNeptuneRunner as Runner elif os.environ.get("USE_ALCHEMY", "0") == "1": from catalyst.dl import SupervisedAlchemyRunner as Runner else: from catalyst.dl import SupervisedRunner as Runner from .models import (TIMMModels, TIMMetricLearningMModels, proxy_model, SSUnet) from .callbacks import (MultiDiceCallback, MacroF2ScoreCallback) from .losses import (LabelSmoothingCrossEntropy, MetricLearningLoss, MultiDiceLoss) registry.MODELS.add_from_module(m) registry.EXPERIMENTS.add_from_module(exp) registry.Model(TIMMModels) registry.Model(TIMMetricLearningMModels) registry.Model(proxy_model) registry.Model(SSUnet) registry.Callback(MultiDiceCallback) registry.Callback(MacroF2ScoreCallback) registry.Criterion(LabelSmoothingCrossEntropy) registry.Criterion(MetricLearningLoss) registry.Criterion(MultiDiceLoss)
from catalyst.dl import SupervisedRunner as Runner from catalyst.dl import registry from src.symmetric_lovasz_loss import SymmetricLovaszLoss from .callbacks import ( CustomDiceCallback, CustomInferCallback, PostprocessingCallback, PseudoLabelsCallback, CheckpointLoader, ) from .experiment import Experiment from .losses import BCEDiceLossCustom registry.Criterion(BCEDiceLossCustom) registry.Criterion(SymmetricLovaszLoss) registry.Callback(CustomDiceCallback) registry.Callback(PostprocessingCallback) registry.Callback(CustomInferCallback) registry.Callback(PseudoLabelsCallback) registry.Callback(CheckpointLoader)
from catalyst.dl import registry from torch_optimizer import Ranger from .callbacks import (CosineLossCallback, KLDivLossCallback, MaskedLanguageModelCallback, MSELossCallback, PerplexityMetricCallbackDistillation, CarbontrackerCallback) from .experiment import Experiment # noqa: F401 from .models import BertForMLM, DistilbertStudentModel from .runners import DistilMLMRunner as Runner # noqa: F401 registry.Model(BertForMLM) registry.Model(DistilbertStudentModel) registry.Optimizer(Ranger) registry.Callback(CosineLossCallback) registry.Callback(MaskedLanguageModelCallback) registry.Callback(KLDivLossCallback) registry.Callback(MSELossCallback) registry.Callback(PerplexityMetricCallbackDistillation) registry.Callback(CarbontrackerCallback)
registry.Model(ResUnet) registry.Model(UNetResNet) registry.Model(LinkNet34) registry.Model(DenseNetDetector) registry.Model(ResnetDetector) registry.Model(resnet34) registry.Model(SCseUnet) registry.Model(ResUnetScSeDecoded) registry.Model(QUnet) registry.Model(EfficientUnet) registry.Model(PretrainedResnet) registry.Model(PretrainedDensenet) registry.Model(ModelFromCheckpoint) registry.Callback(MeanDiceCallback) registry.Callback(AllAccuracyCallback) registry.Callback(F1Callback) registry.Callback(FBetaCallback) registry.Optimizer(PlainRAdam) registry.Optimizer(AdamW) registry.Criterion(JointLoss) registry.Criterion(CCE) registry.Criterion(BinaryDiceLoss) registry.Criterion(BinaryDiceLogLoss) registry.Criterion(MulticlassDiceLoss) registry.Criterion(TverskyLoss) registry.Criterion(DiceAndBCE) registry.Criterion(FocalLossMultiChannel)
from pytorch_toolbelt.losses import FocalLoss from src.models.efficient_new import EfficientNew from .models.multihead import MultiHeadNet from .models.efficient import Efficient from .experiment import Experiment from .losses import * from .callbacks import * # Register models registry.Model(MultiHeadNet) registry.Model(Efficient) registry.Model(EfficientNew) # Register callbacks registry.Callback(HMacroAveragedRecall) registry.Callback(UnFreezeCallback) registry.Callback(FreezeCallback) registry.Callback(ImageViewerCallback) registry.Callback(MixupCutmixCallback) registry.Callback(CheckpointLoader) registry.Callback(HMacroAveragedRecallSingle) # Register criterion registry.Criterion(FocalLoss) registry.Criterion(OHEMLoss) registry.Criterion(ReducedFocalLoss) registry.Criterion(LabelSmoothingLoss)
from catalyst.dl import registry from catalyst.dl import SupervisedRunner as Runner from .experiment import Experiment from utils.callbacks import DiceCallback as MyDice, IouCallback as MyIOU from .model import Model registry.Callback(MyDice, name='MyDice') registry.Callback(MyIOU, name='MyIOU')
from catalyst.dl import registry, SupervisedRunner as Runner from experiment import Experiment from model import ZindiModel from callbacks import LogLoss registry.Model(ZindiModel) registry.Callback(LogLoss)
# Register models registry.Model(Linknet) registry.Model(FPN) registry.Model(PSPNet) registry.Model(Unet) registry.Model(UnetIBN) # registry.Model(model34_DeepSupervion) # registry.Model(HyperUnet) registry.Model(UnetSCSE) registry.Model(Res34Unetv4) registry.Model(UnetMix) registry.Model(UnetOC) registry.Model(LadderNetv6) # Register callbacks registry.Callback(LabelSmoothCriterionCallback) registry.Callback(SmoothMixupCallback) registry.Callback(DSAccuracyCallback) registry.Callback(DSCriterionCallback) registry.Callback(SlackLogger) registry.Callback(DiceCallbackApex) registry.Callback(SIIMCriterionCallback) # Register criterions registry.Criterion(LabelSmoothingCrossEntropy) registry.Criterion(BCEDiceLossApex) registry.Criterion(BCEFocalLossApex) registry.Criterion(WeightedBCE) registry.Criterion(WeightedBCEDiceLossApex) # Register optimizers
# flake8: noqa from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from models import * from callbacks import * from optimizers import * # Register models registry.Model(GAIN) registry.Model(GCAM) # Register callbacks registry.Callback(GAINCriterionCallback) registry.Callback(GAINSaveHeatmapCallback) registry.Callback(GCAMSaveHeatmapCallback) registry.Callback(GAINMaskCriterionCallback) # Register criterions # Register optimizers registry.Optimizer(AdamW) registry.Optimizer(Nadam)
# flake8: noqa # from .runner import Runner from catalyst.dl import SupervisedRunner as Runner from catalyst.dl import registry from .experiment import Experiment from .callbacks import DecoderCallback, MeanAPCallback from .losses import CenterNetDetectionLoss, \ RegL1Loss, MSEIndLoss, BCEIndLoss, FocalIndLoss from . import models registry.Criterion(CenterNetDetectionLoss) registry.Criterion(RegL1Loss) registry.Criterion(MSEIndLoss) registry.Criterion(BCEIndLoss) registry.Criterion(FocalIndLoss) registry.Callback(DecoderCallback) registry.Callback(MeanAPCallback) registry.MODELS.add_from_module(models)
# flake8: noqa # pylint: disable=unused-import from catalyst.dl import registry from transformers import AdamW, WarmupLinearSchedule from .experiment import Experiment from .catalyst_ext.runner import BertSupervisedRunner as Runner from .model_wrapper import BertModel from .catalyst_ext.bert_criterion import BertCrossEntropyLoss, BertCriterionCallback registry.Model(BertModel) registry.Criterion(BertCrossEntropyLoss) registry.Callback(BertCriterionCallback) registry.Optimizer(AdamW, name='TransformersAdamW') registry.Scheduler(WarmupLinearSchedule)
# flake8: noqa from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from .callbacks import * from .models import * from .losses import * registry.Model(Finetune) registry.Callback(F1Callback) registry.Callback(FbetaCallback) registry.Callback(MixupLossCallback) registry.Callback(IterCheckpointCallback) # Register loss registry.Criterion(FocalLoss) registry.Criterion(FbetaLoss) registry.Criterion(BCEAndFbeta) registry.Criterion(BCEFbetaFocalLoss)
# flake8: noqa from .experiment import Experiment from catalyst.dl import registry from catalyst.dl import SupervisedRunner as Runner from src.callbacks.tensorboard import VisualizationCallback, ProjectorCallback from src.callbacks.cico.doe import DoECallback from src.callbacks.cico.benchmark import BenchmarkingCallback from src.models.cico.generic import GenericModel from src.schedulers.cosine import CosineAnnealingWarmUpRestarts from src.losses.cico.arcface import ArcFaceLinear, ArcFaceLoss, L2Norm from src.losses.cico.triplet import TripletSemiHardLoss registry.Model(GenericModel) registry.Module(L2Norm) registry.Module(ArcFaceLinear) registry.Criterion(ArcFaceLoss) registry.Criterion(TripletSemiHardLoss) registry.Callback(VisualizationCallback) registry.Callback(ProjectorCallback) registry.Callback(DoECallback) registry.Callback(BenchmarkingCallback) registry.Scheduler(CosineAnnealingWarmUpRestarts)
from catalyst.dl import registry from .runner import Runner from .experiment import Experiment from .callbacks import CharErrorRateCallback from .optimizers import SWA from .models import ( LightLSTM, DeepSpeech, DeepSpeechV2, LightConv, LookaheadLSTM, ) registry.Callback(CharErrorRateCallback) registry.Model(LightLSTM) registry.Model(DeepSpeech) registry.Model(DeepSpeechV2) registry.Model(LightConv) registry.Model(LookaheadLSTM) registry.Optimizer(SWA)
from src.callbacks import ( DiscriminatorLossCallback, GeneratorLossCallback, GenerateAudioCallback, ShuffleDatasetCallback, ) from src.models import Generator, Discriminator from src.runner import MelGANRunner as Runner from catalyst.dl import registry from src.experiment import Experiment registry.Model(Generator) registry.Model(Discriminator) registry.Callback(GeneratorLossCallback) registry.Callback(DiscriminatorLossCallback) registry.Callback(GenerateAudioCallback) registry.Callback(ShuffleDatasetCallback)
from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from .callbacks import KappaCallback, KappaCriterionCallback, SmoothCCECallback, OrdinalCriterionCallback from .model import resnet34, resnet34_pretrained, resnext50_pretrained, efficientnet_pretrained, cadene_model, \ ordinal_efficientnet registry.Model(resnet34) registry.Model(resnet34_pretrained) registry.Model(resnext50_pretrained) registry.Model(efficientnet_pretrained) registry.Model(cadene_model) registry.Model(ordinal_efficientnet) registry.Callback(KappaCallback) registry.Callback(KappaCriterionCallback) registry.Callback(SmoothCCECallback) registry.Callback(OrdinalCriterionCallback)
import sys import warnings from catalyst.dl import registry from .runner import Runner from .experiment import Experiment from .models import BertBasedMLM from .callbacks import PerplexityCallback if not sys.warnoptions: warnings.simplefilter("ignore") registry.Model(BertBasedMLM) registry.Callback(PerplexityCallback)
from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from models import * from losses import * from callbacks import * from optimizers import * from schedulers import * from segmentation_models_pytorch import Unet as smpUnet # Register models registry.Model(UNet3D) registry.Model(UNet3D2) registry.Model(ResidualUNet3D) registry.Model(VNet) registry.Model(smpUnet) registry.Model(DeepLab) registry.MODELS._late_add_callbacks = [] # Register callbacks registry.Callback(MultiDiceCallback) # Register criterions registry.Criterion(MultiDiceLoss) # Register optimizers # registry.Optimizer(AdamW) # registry.Optimizer(Nadam) # registry.Scheduler(CyclicLRFix)
# flake8: noqa from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from .callbacks import MyLossCallback, IterCheckpointCallback from .models import Net, FewShotModel from .losses import * registry.Model(Net) registry.Model(FewShotModel) registry.Callback(MyLossCallback) registry.Callback(IterCheckpointCallback) registry.Criterion(FocalLoss)
RecallCallback, F1Callback, FBetaCallback, SpearmanScoreCallback, ) from .experiment import Experiment registry.Model(LinearModel) registry.Model(LSTM_GRU) registry.Model(MultiInputLstm) registry.Model(MultiInputLstmGru) registry.Model(MultiInputLstmGruAttention) registry.Model(TransfModel) registry.Model(PooledTransfModel) registry.Model(PooledLstmTransfModel) registry.Model(PooledTransfModelWithCatericalFeatures) registry.Model(PTCFS) registry.Model(PTM) registry.Model(PTC) registry.Model(TwoSidedPooledTransformer) registry.Model(PCTCFS) # functions registry.Model(patch_model_with_embedding) registry.Model(model_from_checkpoint) registry.Model(unfreezed_transf) registry.Callback(PrecisionCallback) registry.Callback(RecallCallback) registry.Callback(F1Callback) registry.Callback(FBetaCallback) registry.Callback(SpearmanScoreCallback)
# flake8: noqa from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from models import * from losses import * from callbacks import * from optimizers import * # Register models registry.Model(CNNFinetuneModels) registry.Model(TIMMModels) registry.Model(MultiModals) # Register callbacks registry.Callback(MultiTaskCriterionCallback) registry.Criterion(LogLoss) registry.Optimizer(Nadam) registry.Optimizer(AdamW)
# flake8: noqa # isort:skip_file from catalyst.dl import registry, SupervisedRunner as Runner from .callbacks import PredictionCallback from .experiment import Experiment from .model import MultiHeadNet from efficientnet_pytorch import EfficientNet from catalyst.contrib.models.cv import ResnetEncoder registry.Model(MultiHeadNet) registry.Model(EfficientNet.from_pretrained, name='EfficientNet') registry.Model(ResnetEncoder) registry.Callback(PredictionCallback)
# flake8: noqa from catalyst.dl import registry from .experiment import Experiment from .runner import ModelRunner as Runner from models import * from losses import * from callbacks import * from optimizers import * # Register models registry.Model(ResNet) registry.Model(cell_senet) registry.Model(cell_densenet) # Register callbacks registry.Callback(LabelSmoothCriterionCallback) # Register criterions registry.Criterion(LabelSmoothingCrossEntropy) # Register optimizers registry.Optimizer(AdamW) registry.Optimizer(Nadam) registry.Optimizer(RAdam)
from catalyst.dl import registry from catalyst.dl import SupervisedRunner as Runner from .models import SimpleNet from .callbacks import DoSomethingWithDataCallback from .experiment import Experiment registry.Model(SimpleNet) registry.Callback(DoSomethingWithDataCallback)
from .metrics import ( WeightedAUC, # for binary classification SingleClassWeightedAUC # for multiclass classification ) from .schedulers import CosineAnnealingWithRestartsLR from .criterions import LabelSmoothingLoss from .models import ( BinaryEfficientNet, MulticlassEfficientNet, StemMulticlassEfficientNet, BinaryDensenet, LLFEfficientNet, patch_efficientnet_backbone, patch_efficientnet_conv_stem, ) registry.Callback(WeightedAUC) registry.Callback(SingleClassWeightedAUC) registry.Scheduler(CosineAnnealingWithRestartsLR) registry.Criterion(LabelSmoothingLoss) registry.Model(BinaryEfficientNet) registry.Model(MulticlassEfficientNet) registry.Model(StemMulticlassEfficientNet) registry.Model(BinaryDensenet) registry.Model(LLFEfficientNet) registry.Model(patch_efficientnet_backbone) registry.Model(patch_efficientnet_conv_stem)