from torch.optim.lr_scheduler import (
CosineAnnealingLR,
CosineAnnealingWarmRestarts,
CyclicLR,
ReduceLROnPlateau,
StepLR,
)
from src.utils.mapper import configmapper
configmapper.map("schedulers", "step")(StepLR)
configmapper.map("schedulers", "cosineanneal")(CosineAnnealingLR)
configmapper.map("schedulers", "reduceplateau")(ReduceLROnPlateau)
configmapper.map("schedulers", "cyclic")(CyclicLR)
configmapper.map("schedulers",
"cosineannealrestart")(CosineAnnealingWarmRestarts)
"""Metrics."""
from sklearn.metrics import (
mean_squared_error,
f1_score,
precision_score,
recall_score,
roc_auc_score,
accuracy_score,
)
from src.utils.mapper import configmapper
configmapper.map("metrics", "sklearn_f1")(f1_score)
configmapper.map("metrics", "sklearn_p")(precision_score)
configmapper.map("metrics", "sklearn_r")(recall_score)
configmapper.map("metrics", "sklearn_roc")(roc_auc_score)
configmapper.map("metrics", "sklearn_acc")(accuracy_score)
configmapper.map("metrics", "sklearn_mse")(mean_squared_error)
from transformers import AutoModelForTokenClassification, AutoModelForQuestionAnswering
from src.utils.mapper import configmapper
configmapper.map("models", "autotoken")(AutoModelForTokenClassification)
configmapper.map("models", "autospans")(AutoModelForQuestionAnswering)
import torch.nn as nn
from src.utils.mapper import configmapper
configmapper.map("activations", "relu")(nn.ReLU)
configmapper.map("activations", "logsoftmax")(nn.LogSoftmax)
configmapper.map("activations", "softmax")(nn.Softmax)
import matplotlib.pyplot as plt
import numpy as np
import torch
import torch_geometric.transforms as torch_geometric_transforms
import torchvision.transforms as transforms
from skimage.segmentation import slic
from torch_geometric.data import Data
from torch_scatter import scatter_mean
from src.utils.mapper import configmapper
configmapper.map("transforms", "Resize")(transforms.Resize)
configmapper.map("transforms", "Normalize")(transforms.Normalize)
configmapper.map("transforms", "ToTensor")(transforms.ToTensor)
configmapper.map("transforms", "ToPILImage")(transforms.ToPILImage)
configmapper.map("transforms", "Grayscale")(transforms.Grayscale)
configmapper.map("transforms",
"ToSLIC")(torch_geometric_transforms.to_superpixels.ToSLIC)
configmapper.map("transforms", "KNNGraph")(torch_geometric_transforms.KNNGraph)
configmapper.map("transforms",
"PolarTransformation")(torch_geometric_transforms.Polar)
configmapper.map("transforms",
"RadiusGraph")(torch_geometric_transforms.RadiusGraph)
# https://github.com/phcavelar/mnist-superpixel
@configmapper.map("transforms", "RAGGraph")
class RAGGraph(object):
def __init__(self, add_seg=False, add_img=False, **kwargs):
self.add_seg = add_seg
self.add_img = add_img
"All criterion functions."
from torch.nn import MSELoss, CrossEntropyLoss
from src.utils.mapper import configmapper
configmapper.map("losses", "mse")(MSELoss)
configmapper.map("losses", "CrossEntropyLoss")(CrossEntropyLoss)
" Method containing activation functions"
from torch.optim import SGD, Adam, AdamW
from src.utils.mapper import configmapper
configmapper.map("optimizers", "adam")(Adam)
configmapper.map("optimizers", "adam_w")(AdamW)
configmapper.map("optimizers", "sgd")(SGD)