masks_augments_compare-21/util/module_mixins.py
2020-12-17 08:02:29 +01:00

417 lines
17 KiB
Python

# Imports from python Internals
from abc import ABC
from itertools import cycle
from collections import defaultdict, namedtuple
# Numerical Imports, Metrics and Plotting
import numpy as np
from sklearn.ensemble import IsolationForest
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay, roc_auc_score, roc_curve, auc, f1_score, \
recall_score, average_precision_score
from matplotlib import pyplot as plt
# Import Deep Learning Framework
import torch
from torch import nn
from torch.optim import Adam
from torch.optim.lr_scheduler import CosineAnnealingWarmRestarts
from torch.utils.data import DataLoader
from torchcontrib.optim import SWA
from torchvision.transforms import Compose, RandomApply
# Import Functions and Modules from MLLIB
from ml_lib.audio_toolset.mel_augmentation import NoiseInjection, LoudnessManipulator, ShiftTime, MaskAug
from ml_lib.audio_toolset.audio_io import NormalizeLocal
from ml_lib.modules.util import LightningBaseModule
from ml_lib.utils.tools import to_one_hot
from ml_lib.utils.transforms import ToTensor
# Import Project Variables
import variables as V
class BaseLossMixin:
absolute_loss = nn.L1Loss()
nll_loss = nn.NLLLoss()
bce_loss = nn.BCELoss()
ce_loss = nn.CrossEntropyLoss()
class BaseOptimizerMixin:
def configure_optimizers(self):
assert isinstance(self, LightningBaseModule)
optimizer_dict = dict(
# 'optimizer':optimizer, # The Optimizer
# 'lr_scheduler': scheduler, # The LR scheduler
frequency=1, # The frequency of the scheduler
interval='epoch', # The unit of the scheduler's step size
# 'reduce_on_plateau': False, # For ReduceLROnPlateau scheduler
# 'monitor': 'mean_val_loss' # Metric to monitor
)
optimizer = Adam(params=self.parameters(), lr=self.params.lr, weight_decay=self.params.weight_decay)
if self.params.sto_weight_avg:
optimizer = SWA(optimizer, swa_start=10, swa_freq=5, swa_lr=0.05)
optimizer_dict.update(optimizer=optimizer)
if self.params.lr_warmup_steps:
scheduler = CosineAnnealingWarmRestarts(optimizer, self.params.lr_warmup_steps)
optimizer_dict.update(lr_scheduler=scheduler)
return optimizer_dict
def on_train_end(self):
assert isinstance(self, LightningBaseModule)
for opt in self.trainer.optimizers:
if isinstance(opt, SWA):
opt.swap_swa_sgd()
def on_epoch_end(self):
assert isinstance(self, LightningBaseModule)
if self.params.opt_reset_interval:
if self.current_epoch % self.params.opt_reset_interval == 0:
for opt in self.trainer.optimizers:
opt.state = defaultdict(dict)
class BaseTrainMixin:
def training_step(self, batch_xy, batch_nb, *args, **kwargs):
assert isinstance(self, LightningBaseModule)
batch_x, batch_y = batch_xy
y = self(batch_x).main_out
loss = self.ce_loss(y.squeeze(), batch_y.long())
return dict(loss=loss)
def training_epoch_end(self, outputs):
assert isinstance(self, LightningBaseModule)
keys = list(outputs[0].keys())
summary_dict = {f'mean_{key}': torch.mean(torch.stack([output[key]
for output in outputs]))
for key in keys if 'loss' in key}
for key in summary_dict.keys():
self.log(key, summary_dict[key])
class BaseValMixin:
def validation_step(self, batch_xy, batch_idx, *args, **kwargs):
assert isinstance(self, LightningBaseModule)
batch_x, batch_y = batch_xy
y = self(batch_x).main_out
val_loss = self.ce_loss(y.squeeze(), batch_y.long())
return dict(val_loss=val_loss,
batch_idx=batch_idx, y=y, batch_y=batch_y)
def validation_epoch_end(self, outputs, *_, **__):
assert isinstance(self, LightningBaseModule)
summary_dict = dict()
keys = list(outputs[0].keys())
summary_dict.update({f'mean_{key}': torch.mean(torch.stack([output[key]
for output in outputs]))
for key in keys if 'loss' in key}
)
additional_scores = self.additional_scores(outputs)
summary_dict.update(**additional_scores)
for key in summary_dict.keys():
self.log(key, summary_dict[key])
class BaseTestMixin:
def test_step(self, batch_xy, batch_idx, *_, **__):
assert isinstance(self, LightningBaseModule)
batch_x, batch_y = batch_xy
y = self(batch_x).main_out
test_loss = self.ce_loss(y.squeeze(), batch_y.long())
return dict(test_loss=test_loss,
batch_idx=batch_idx, y=y, batch_y=batch_y)
def test_epoch_end(self, outputs, *_, **__):
assert isinstance(self, LightningBaseModule)
summary_dict = dict()
keys = list(outputs[0].keys())
summary_dict.update({f'mean_{key}': torch.mean(torch.stack([output[key]
for output in outputs]))
for key in keys if 'loss' in key}
)
additional_scores = self.additional_scores(outputs)
summary_dict.update(**additional_scores)
for key in summary_dict.keys():
self.log(key, summary_dict[key])
class DatasetMixin:
def build_dataset(self):
assert isinstance(self, LightningBaseModule)
# Dataset
# =============================================================================
# Mel Transforms
mel_kwargs = dict(sample_rate=self.params.sr,
n_mels=self.params.n_mels,
n_fft=self.params.n_fft,
hop_length=self.params.hop_length)
# Utility
normalize = NormalizeLocal()
# Data Augmentations
mel_augmentations = Compose([
RandomApply([
NoiseInjection(0.2),
LoudnessManipulator(0.5),
ShiftTime(0.4),
MaskAug(0.2),
], p=0.6),
normalize])
# Datasets
Dataset = namedtuple('Datasets', 'train_dataset val_dataset test_dataset')
dataset = Dataset(self.dataset_class(data_root=self.params.root, # TRAIN DATASET
setting=V.DATA_OPTION_train,
fold=list(range(1,8)),
reset=self.params.reset,
mel_kwargs=mel_kwargs,
mel_augmentations=mel_augmentations),
val_dataset=self.dataset_class(data_root=self.params.root, # VALIDATION DATASET
setting=V.DATA_OPTION_devel,
fold=9,
reset=self.params.reset,
mel_kwargs=mel_kwargs,
mel_augmentations=normalize),
test_dataset=self.dataset_class(data_root=self.params.root, # TEST DATASET
setting=V.DATA_OPTION_test,
fold=10,
reset=self.params.reset,
mel_kwargs=mel_kwargs,
mel_augmentations=normalize),
)
if dataset.train_dataset.task_type == V.TASK_OPTION_binary:
# noinspection PyAttributeOutsideInit
self.additional_scores = BinaryScores(self)
elif dataset.train_dataset.task_type == V.TASK_OPTION_multiclass:
# noinspection PyAttributeOutsideInit
self.additional_scores = MultiClassScores(self)
else:
raise ValueError
return dataset
class BaseDataloadersMixin(ABC):
# Dataloaders
# ================================================================================
# Train Dataloader
def train_dataloader(self):
assert isinstance(self, LightningBaseModule)
# sampler = RandomSampler(self.dataset.train_dataset, True, len(self.dataset.train_dataset))
sampler = None
return DataLoader(dataset=self.dataset.train_dataset, shuffle=True if not sampler else None, sampler=sampler,
batch_size=self.params.batch_size, pin_memory=True,
num_workers=self.params.worker)
# Test Dataloader
def test_dataloader(self):
assert isinstance(self, LightningBaseModule)
return DataLoader(dataset=self.dataset.test_dataset, shuffle=False,
batch_size=self.params.batch_size, pin_memory=True,
num_workers=self.params.worker)
# Validation Dataloader
def val_dataloader(self):
assert isinstance(self, LightningBaseModule)
return DataLoader(dataset=self.dataset.val_dataset, shuffle=False, pin_memory=True,
batch_size=self.params.batch_size, num_workers=self.params.worker)
class BaseScores(ABC):
def __init__(self, lightning_model):
self.model = lightning_model
pass
def __call__(self, outputs):
# summary_dict = dict()
# return summary_dict
raise NotImplementedError
class MultiClassScores(BaseScores):
def __init__(self, *args):
super(MultiClassScores, self).__init__(*args)
pass
def __call__(self, outputs):
summary_dict = dict()
#######################################################################################
# Additional Score - UAR - ROC - Conf. Matrix - F1
#######################################################################################
#
# INIT
y_true = torch.cat([output['batch_y'] for output in outputs]).cpu().numpy()
y_true_one_hot = to_one_hot(y_true, self.model.n_classes)
y_pred = torch.cat([output['y'] for output in outputs]).squeeze().cpu().float().numpy()
y_pred_max = np.argmax(y_pred, axis=1)
class_names = {val: key for key, val in self.model.dataset.test_dataset.classes.items()}
######################################################################################
#
# F1 SCORE
micro_f1_score = f1_score(y_true, y_pred_max, labels=None, pos_label=1, average='micro', sample_weight=None,
zero_division=True)
macro_f1_score = f1_score(y_true, y_pred_max, labels=None, pos_label=1, average='macro', sample_weight=None,
zero_division=True)
summary_dict.update(dict(micro_f1_score=micro_f1_score, macro_f1_score=macro_f1_score))
#######################################################################################
#
# ROC Curve
# Compute ROC curve and ROC area for each class
fpr = dict()
tpr = dict()
roc_auc = dict()
for i in range(self.model.n_classes):
fpr[i], tpr[i], _ = roc_curve(y_true_one_hot[:, i], y_pred[:, i])
roc_auc[i] = auc(fpr[i], tpr[i])
# Compute micro-average ROC curve and ROC area
fpr["micro"], tpr["micro"], _ = roc_curve(y_true_one_hot.ravel(), y_pred.ravel())
roc_auc["micro"] = auc(fpr["micro"], tpr["micro"])
# First aggregate all false positive rates
all_fpr = np.unique(np.concatenate([fpr[i] for i in range(self.model.n_classes)]))
# Then interpolate all ROC curves at this points
mean_tpr = np.zeros_like(all_fpr)
for i in range(self.model.n_classes):
mean_tpr += np.interp(all_fpr, fpr[i], tpr[i])
# Finally average it and compute AUC
mean_tpr /= self.model.n_classes
fpr["macro"] = all_fpr
tpr["macro"] = mean_tpr
roc_auc["macro"] = auc(fpr["macro"], tpr["macro"])
# Plot all ROC curves
plt.figure()
plt.plot(fpr["micro"], tpr["micro"],
label=f'micro ROC ({round(roc_auc["micro"], 2)})',
color='deeppink', linestyle=':', linewidth=4)
plt.plot(fpr["macro"], tpr["macro"],
label=f'macro ROC({round(roc_auc["macro"], 2)})',
color='navy', linestyle=':', linewidth=4)
colors = cycle(['firebrick', 'orangered', 'gold', 'olive', 'limegreen', 'aqua',
'dodgerblue', 'slategrey', 'royalblue', 'indigo', 'fuchsia'], )
for i, color in zip(range(self.model.n_classes), colors):
plt.plot(fpr[i], tpr[i], color=color, lw=2, label=f'{class_names[i]} ({round(roc_auc[i], 2)})')
plt.plot([0, 1], [0, 1], 'k--', lw=2)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.legend(loc="lower right")
self.model.logger.log_image('ROC', image=plt.gcf(), step=self.model.current_epoch)
self.model.logger.log_image('ROC', image=plt.gcf(), step=self.model.current_epoch, ext='pdf')
plt.clf()
#######################################################################################
#
# ROC SCORE
try:
macro_roc_auc_ovr = roc_auc_score(y_true_one_hot, y_pred, multi_class="ovr",
average="macro")
summary_dict.update(macro_roc_auc_ovr=macro_roc_auc_ovr)
except ValueError:
micro_roc_auc_ovr = roc_auc_score(y_true_one_hot, y_pred, multi_class="ovr",
average="micro")
summary_dict.update(micro_roc_auc_ovr=micro_roc_auc_ovr)
#######################################################################################
#
# Confusion matrix
cm = confusion_matrix([class_names[x] for x in y_true], [class_names[x] for x in y_pred_max],
labels=[class_names[key] for key in class_names.keys()],
normalize='all')
disp = ConfusionMatrixDisplay(confusion_matrix=cm,
display_labels=[class_names[i] for i in range(self.model.n_classes)]
)
disp.plot(include_values=True)
self.model.logger.log_image('Confusion_Matrix', image=disp.figure_, step=self.model.current_epoch)
self.model.logger.log_image('Confusion_Matrix', image=disp.figure_, step=self.model.current_epoch, ext='pdf')
plt.close('all')
return summary_dict
class BinaryScores(BaseScores):
def __init__(self, *args):
super(BinaryScores, self).__init__(*args)
def __call__(self, outputs):
summary_dict = dict()
# Additional Score like the unweighted Average Recall:
#########################
# UnweightedAverageRecall
y_true = torch.cat([output['batch_y'] for output in outputs]) .cpu().numpy()
y_pred = torch.cat([output['element_wise_recon_error'] for output in outputs]).squeeze().cpu().numpy()
# How to apply a threshold manualy
# y_pred = (y_pred >= 0.5).astype(np.float32)
# How to apply a threshold by IF (Isolation Forest)
clf = IsolationForest(random_state=self.model.seed)
y_score = clf.fit_predict(y_pred.reshape(-1,1))
y_score = (np.asarray(y_score) == -1).astype(np.float32)
uar_score = recall_score(y_true, y_score, labels=[0, 1], average='macro',
sample_weight=None, zero_division='warn')
summary_dict.update(dict(uar_score=uar_score))
#########################
# Precission
precision_score = average_precision_score(y_true, y_score)
summary_dict.update(dict(precision_score=precision_score))
#########################
# AUC
try:
auc_score = roc_auc_score(y_true=y_true, y_score=y_score)
summary_dict.update(dict(auc_score=auc_score))
except ValueError:
summary_dict.update(dict(auc_score=-1))
#########################
# pAUC
try:
pauc = roc_auc_score(y_true=y_true, y_score=y_score, max_fpr=0.15)
summary_dict.update(dict(pauc_score=pauc))
except ValueError:
summary_dict.update(dict(pauc_score=-1))
return summary_dict