From e7d1a4895ad4bdc26cacca046f1d8dd55d1aa2b6 Mon Sep 17 00:00:00 2001 From: Si11ium Date: Thu, 14 May 2020 23:08:36 +0200 Subject: [PATCH] requirements --- _paramters.py | 16 +++++++-------- datasets/binar_masks.py | 7 ++++--- main_inference.py | 21 +++++++++++++++----- models/bandwise_conv_classifier.py | 4 ++-- models/bandwise_conv_multihead_classifier.py | 14 ++++++------- models/conv_classifier.py | 4 ++-- models/ensemble.py | 16 +++++++-------- models/residual_conv_classifier.py | 6 ++++-- util/module_mixins.py | 2 +- 9 files changed, 52 insertions(+), 38 deletions(-) diff --git a/_paramters.py b/_paramters.py index f707231..9d54918 100644 --- a/_paramters.py +++ b/_paramters.py @@ -30,12 +30,12 @@ main_arg_parser.add_argument("--data_n_fft", type=int, default=512, help="") main_arg_parser.add_argument("--data_mixup", type=strtobool, default=False, help="") # Transformation Parameters -main_arg_parser.add_argument("--data_loudness_ratio", type=float, default=0.4, help="") -main_arg_parser.add_argument("--data_shift_ratio", type=float, default=0.3, help="") -main_arg_parser.add_argument("--data_noise_ratio", type=float, default=0.4, help="") -main_arg_parser.add_argument("--data_mask_ratio", type=float, default=0.2, help="") -main_arg_parser.add_argument("--data_speed_ratio", type=float, default=0.3, help="") -main_arg_parser.add_argument("--data_speed_factor", type=float, default=0.7, help="") +main_arg_parser.add_argument("--data_loudness_ratio", type=float, default=0, help="") # 0.4 +main_arg_parser.add_argument("--data_shift_ratio", type=float, default=0, help="") # 0.3 +main_arg_parser.add_argument("--data_noise_ratio", type=float, default=0, help="") # 0.4 +main_arg_parser.add_argument("--data_mask_ratio", type=float, default=0.2, help="") # 0.2 +main_arg_parser.add_argument("--data_speed_ratio", type=float, default=0.3, help="") # 0.3 +main_arg_parser.add_argument("--data_speed_factor", type=float, default=0.7, help="") # 0.7 # Training Parameters main_arg_parser.add_argument("--train_outpath", type=str, default="output", help="") @@ -49,8 +49,8 @@ main_arg_parser.add_argument("--train_lr", type=float, default=1e-4, help="") main_arg_parser.add_argument("--train_num_sanity_val_steps", type=int, default=0, help="") # Model Parameters -main_arg_parser.add_argument("--model_type", type=str, default="CC", help="") -main_arg_parser.add_argument("--model_secondary_type", type=str, default="CC", help="") +main_arg_parser.add_argument("--model_type", type=str, default="BCMC", help="") +main_arg_parser.add_argument("--model_secondary_type", type=str, default="BCMC", help="") main_arg_parser.add_argument("--model_weight_init", type=str, default="xavier_normal_", help="") main_arg_parser.add_argument("--model_activation", type=str, default="leaky_relu", help="") main_arg_parser.add_argument("--model_filters", type=str, default="[32, 64, 128, 64]", help="") diff --git a/datasets/binar_masks.py b/datasets/binar_masks.py index 5c03a7c..50fdb6c 100644 --- a/datasets/binar_masks.py +++ b/datasets/binar_masks.py @@ -47,9 +47,10 @@ class BinaryMasksDataset(Dataset): filename, label = row.strip().split(',') labeldict[filename] = self._to_label[label.lower()] if not self.setting == 'test' else filename if self.stretch and self.setting == V.DATA_OPTIONS.train: - labeldict.update({f'X_{key}': val for key, val in labeldict.items()}) - labeldict.update({f'X_X_{key}': val for key, val in labeldict.items()}) - labeldict.update({f'X_X_X_{key}': val for key, val in labeldict.items()}) + additional_dict = ({f'X_{key}': val for key, val in labeldict.items()}) + additional_dict.update({f'X_X_{key}': val for key, val in labeldict.items()}) + additional_dict.update({f'X_X_X_{key}': val for key, val in labeldict.items()}) + labeldict.update(additional_dict) return labeldict def __len__(self): diff --git a/main_inference.py b/main_inference.py index 825082d..2480f14 100644 --- a/main_inference.py +++ b/main_inference.py @@ -5,7 +5,7 @@ from tqdm import tqdm import variables as V from torch.utils.data import DataLoader, Dataset -from torchvision.transforms import Compose +from torchvision.transforms import Compose, RandomApply from ml_lib.audio_toolset.audio_io import AudioToMel, NormalizeLocal, MelToImage @@ -13,6 +13,7 @@ from ml_lib.audio_toolset.audio_io import AudioToMel, NormalizeLocal, MelToImage # ============================================================================= # Transforms +from ml_lib.audio_toolset.mel_augmentation import NoiseInjection, LoudnessManipulator, ShiftTime, MaskAug from ml_lib.utils.logging import Logger from ml_lib.utils.model_io import SavedLightningModels from ml_lib.utils.transforms import ToTensor @@ -28,8 +29,18 @@ def prepare_dataloader(config_obj): AudioToMel(sr=config_obj.data.sr, n_mels=config_obj.data.n_mels, n_fft=config_obj.data.n_fft, hop_length=config_obj.data.hop_length), MelToImage()]) transforms = Compose([NormalizeLocal(), ToTensor()]) + aug_transforms = Compose([ + RandomApply([ + NoiseInjection(config_obj.data.noise_ratio), + LoudnessManipulator(config_obj.data.loudness_ratio), + ShiftTime(config_obj.data.shift_ratio), + MaskAug(config_obj.data.mask_ratio), + ], p=0.6), + # Utility + NormalizeLocal(), ToTensor() + ]) - dataset: Dataset = BinaryMasksDataset(config_obj.data.root, setting='test', + dataset: Dataset = BinaryMasksDataset(config_obj.data.root, setting='train', mel_transforms=mel_transforms, transforms=transforms ) # noinspection PyTypeChecker @@ -49,9 +60,9 @@ def restore_logger_and_model(config_obj): if __name__ == '__main__': outpath = Path('output') - model_type = 'BandwiseConvMultiheadClassifier' - parameters = 'BCMC_9c70168a5711c269b33701f1650adfb9/' - version = 'version_1' + model_type = 'CC' + parameters = 'CC_213adb16e46592c5a405abfbd693835e/' + version = 'version_41' config_filename = 'config.ini' inference_out = 'manual_test_out.csv' diff --git a/models/bandwise_conv_classifier.py b/models/bandwise_conv_classifier.py index 1a1ba43..b54000d 100644 --- a/models/bandwise_conv_classifier.py +++ b/models/bandwise_conv_classifier.py @@ -5,11 +5,11 @@ from torch.nn import ModuleList from ml_lib.modules.blocks import ConvModule, LinearModule from ml_lib.modules.utils import (LightningBaseModule, HorizontalSplitter, HorizontalMerger) -from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetFunction, +from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetMixin, BaseDataloadersMixin) -class BandwiseConvClassifier(BinaryMaskDatasetFunction, +class BandwiseConvClassifier(BinaryMaskDatasetMixin, BaseDataloadersMixin, BaseTrainMixin, BaseValMixin, diff --git a/models/bandwise_conv_multihead_classifier.py b/models/bandwise_conv_multihead_classifier.py index d104a86..7b5741f 100644 --- a/models/bandwise_conv_multihead_classifier.py +++ b/models/bandwise_conv_multihead_classifier.py @@ -6,11 +6,11 @@ from torch.nn import ModuleList from ml_lib.modules.blocks import ConvModule, LinearModule from ml_lib.modules.utils import (LightningBaseModule, Flatten, HorizontalSplitter) -from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetFunction, +from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetMixin, BaseDataloadersMixin) -class BandwiseConvMultiheadClassifier(BinaryMaskDatasetFunction, +class BandwiseConvMultiheadClassifier(BinaryMaskDatasetMixin, BaseDataloadersMixin, BaseTrainMixin, BaseValMixin, @@ -42,7 +42,7 @@ class BandwiseConvMultiheadClassifier(BinaryMaskDatasetFunction, return_dict = {f'band_{band_idx}_val_loss': band_y for band_idx, band_y in enumerate(bands_y_losses)} last_bce_loss = self.bce_loss(y, batch_y) - return_dict.update(last_bce_loss=last_bce_loss) + return_dict.update(last_val_bce_loss=last_bce_loss) bands_y_losses.append(last_bce_loss) combined_loss = torch.stack(bands_y_losses).mean() @@ -76,7 +76,7 @@ class BandwiseConvMultiheadClassifier(BinaryMaskDatasetFunction, last_shape = self.split.shape conv_list = ModuleList() for filters in self.conv_filters: - conv_list.append(ConvModule(last_shape, filters, (k,k), conv_stride=(1, 1), + conv_list.append(ConvModule(last_shape, filters, (k, k), conv_stride=(2, 2), conv_padding=2, **self.params.module_kwargs)) last_shape = conv_list[-1].shape # self.conv_list.append(ConvModule(last_shape, 1, 1, conv_stride=1, **self.params.module_kwargs)) @@ -84,10 +84,10 @@ class BandwiseConvMultiheadClassifier(BinaryMaskDatasetFunction, self.band_list.append(conv_list) self.bandwise_deep_list_1 = ModuleList([ - LinearModule(self.band_list[0][-1].shape, self.params.lat_dim * 4, **self.params.module_kwargs) + LinearModule(self.band_list[0][-1].shape, self.params.lat_dim, **self.params.module_kwargs) for _ in range(self.n_band_sections)]) self.bandwise_deep_list_2 = ModuleList([ - LinearModule(self.params.lat_dim * 4, self.params.lat_dim * 2, **self.params.module_kwargs) + LinearModule(self.params.lat_dim, self.params.lat_dim * 2, **self.params.module_kwargs) for _ in range(self.n_band_sections)]) self.bandwise_latent_list = ModuleList([ LinearModule(self.params.lat_dim * 2, self.params.lat_dim, **self.params.module_kwargs) @@ -96,7 +96,7 @@ class BandwiseConvMultiheadClassifier(BinaryMaskDatasetFunction, LinearModule(self.params.lat_dim, 1, bias=self.params.bias, activation=nn.Sigmoid) for _ in range(self.n_band_sections)]) - self.full_1 = LinearModule(self.n_band_sections, self.params.lat_dim * 4, **self.params.module_kwargs) + self.full_1 = LinearModule(self.n_band_sections, self.params.lat_dim, **self.params.module_kwargs) self.full_2 = LinearModule(self.full_1.shape, self.params.lat_dim * 2, **self.params.module_kwargs) self.full_3 = LinearModule(self.full_2.shape, self.params.lat_dim, **self.params.module_kwargs) self.full_out = LinearModule(self.full_3.shape, 1, bias=self.params.bias, activation=nn.Sigmoid) diff --git a/models/conv_classifier.py b/models/conv_classifier.py index f9a3210..4fac038 100644 --- a/models/conv_classifier.py +++ b/models/conv_classifier.py @@ -5,11 +5,11 @@ from torch.nn import ModuleList from ml_lib.modules.blocks import ConvModule, LinearModule from ml_lib.modules.utils import LightningBaseModule -from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetFunction, +from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetMixin, BaseDataloadersMixin) -class ConvClassifier(BinaryMaskDatasetFunction, +class ConvClassifier(BinaryMaskDatasetMixin, BaseDataloadersMixin, BaseTrainMixin, BaseValMixin, diff --git a/models/ensemble.py b/models/ensemble.py index dde907b..df41657 100644 --- a/models/ensemble.py +++ b/models/ensemble.py @@ -8,17 +8,17 @@ from torch.nn import ModuleList from ml_lib.modules.utils import LightningBaseModule from ml_lib.utils.config import Config from ml_lib.utils.model_io import SavedLightningModels -from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetFunction, +from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetMixin, BaseDataloadersMixin) -class Ensemble(BinaryMaskDatasetFunction, - BaseDataloadersMixin, - BaseTrainMixin, - BaseValMixin, - BaseOptimizerMixin, - LightningBaseModule - ): +class Ensemble(BinaryMaskDatasetMixin, + BaseDataloadersMixin, + BaseTrainMixin, + BaseValMixin, + BaseOptimizerMixin, + LightningBaseModule + ): def __init__(self, hparams): super(Ensemble, self).__init__(hparams) diff --git a/models/residual_conv_classifier.py b/models/residual_conv_classifier.py index de8209b..b377074 100644 --- a/models/residual_conv_classifier.py +++ b/models/residual_conv_classifier.py @@ -5,11 +5,11 @@ from torch.nn import ModuleList from ml_lib.modules.blocks import ConvModule, LinearModule, ResidualModule from ml_lib.modules.utils import LightningBaseModule -from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetFunction, +from util.module_mixins import (BaseOptimizerMixin, BaseTrainMixin, BaseValMixin, BinaryMaskDatasetMixin, BaseDataloadersMixin) -class ResidualConvClassifier(BinaryMaskDatasetFunction, +class ResidualConvClassifier(BinaryMaskDatasetMixin, BaseDataloadersMixin, BaseTrainMixin, BaseValMixin, @@ -45,6 +45,8 @@ class ResidualConvClassifier(BinaryMaskDatasetFunction, last_shape = self.conv_list[-1].shape self.conv_list.append(ConvModule(last_shape, filters, (k, k), conv_stride=(2, 2), conv_padding=2, **self.params.module_kwargs)) + for param in self.conv_list[-1].parameters(): + param.requires_grad = False last_shape = self.conv_list[-1].shape self.full_1 = LinearModule(self.conv_list[-1].shape, self.params.lat_dim, **self.params.module_kwargs) diff --git a/util/module_mixins.py b/util/module_mixins.py index aa233ba..ea4e2dc 100644 --- a/util/module_mixins.py +++ b/util/module_mixins.py @@ -105,7 +105,7 @@ class BaseValMixin: return summary_dict -class BinaryMaskDatasetFunction: +class BinaryMaskDatasetMixin: def build_dataset(self): assert isinstance(self, LightningBaseModule)