Join the 1-day Testing & QA Summit featuring 15+ Expert Speakers.Register for FREE! Join TestMu Conference

How to use do_checkpoint method in localstack

Best Python code snippet using localstack_python

train.py

Source:train.py

1if __name__ == "__main__":2    import argparse3    from src.IdentificationNetwork import IdentificationNetwork4    from src.TrackingNetwork import TrackingNetwork5    from src.Util import get_fixed_path6    import tensorflow.keras.backend as K7    import src.nets89    parser = argparse.ArgumentParser()1011    parser.add_argument("-i", "--datasets_folder", help="Location of the \"Datasets\" folder", type=str, required=True)12    parser.add_argument("-o", "--save_folder", help="Where to save the models", type=str, required=True)1314    parser.add_argument("-l", "--load_folder", help="If and from where to load the models' weights", type=str,15                        required=False, default=None)1617    parser.add_argument("-pte", "--pre_tracking_epochs", help="Epochs for the pre tracking network", type=int,18                        required=False, default=20)19    parser.add_argument("-te", "--tracking_epochs", help="Epochs for the tracking network", type=int,20                        required=False, default=50)21    parser.add_argument("-ie", "--identification_epochs", help="Epochs for the identification network", type=int,22                        required=False, default=25)2324    parser.add_argument("-spe", "--steps_per_epoch", help="Steps per epoch", type=int, required=False, default=128)25    parser.add_argument("-bs", "--batch_size", help="Batch size", type=int, required=False, default=2)2627    parser.add_argument("-dc", "--do_checkpoint", help="Should save checkpoints", type=bool, required=False, default=True)2829    args = parser.parse_args()3031    datasets_folder = get_fixed_path(args.datasets_folder, replace_backslash=True, add_backslash=True)32    save_folder = get_fixed_path(args.save_folder, replace_backslash=True, add_backslash=True)33    load_folder = get_fixed_path(args.load_folder, replace_backslash=True, add_backslash=True) if args.load_folder is not None else None3435    ####################################################################################################################36    # Pre-tracking network37    ####################################################################################################################38    if args.pre_tracking_epochs > 0:39        K.clear_session()4041        if load_folder is None:42            net_pre_tracking = IdentificationNetwork(src.nets.generate_pre_tracking_model(input_shape=(416, 416, 1), number_of_classes=7))4344            net_pre_tracking.train_model(epochs=args.pre_tracking_epochs,45                                         steps_per_epoch=args.steps_per_epoch,46                                         weights_save_path=save_folder + "pre_tracking.h5",47                                         weights_load_path=None,48                                         do_checkpoint=args.do_checkpoint,49                                         generator=IdentificationNetwork.generator(50                                             folder_images=datasets_folder + "PreTracking/",51                                             batch_size=args.batch_size,52                                             image_ext=(".jpg", ".png")53                                         ))5455    ####################################################################################################################56    # Tracking network57    ####################################################################################################################58    if args.tracking_epochs > 0:59        K.clear_session()6061        net_tracking = TrackingNetwork(src.nets.generate_tracking_model(input_shape=(416, 416, 1),62                                                                        output_shape=(13, 13, 11),63                                                                        number_of_classes=6))6465        load_path_tracking = load_folder + "tracking.h5" if load_folder is not None else save_folder + "pre_tracking.h5"6667        net_tracking.get_model().summary()6869        net_tracking.train_model(epochs=args.tracking_epochs,70                                 steps_per_epoch=args.steps_per_epoch,71                                 weights_save_path=save_folder + "tracking.h5",72                                 weights_load_path=load_path_tracking,73                                 do_checkpoint=args.do_checkpoint,74                                 generator=TrackingNetwork.generator(75                                     folder_images=datasets_folder + "Tracking/",76                                     folder_grids=datasets_folder + "Tracking/",77                                     batch_size=args.batch_size78                                 ))7980    ####################################################################################################################81    # Identification network82    ####################################################################################################################83    if args.identification_epochs > 0:84        K.clear_session()8586        net_identification = IdentificationNetwork(models=(87            src.nets.generate_identification_model(input_shape=(50, 50, 1), number_of_classes=9),88            src.nets.generate_identification_model(input_shape=(50, 50, 1), number_of_classes=4),89            src.nets.generate_identification_model(input_shape=(50, 50, 1), number_of_classes=2),90            src.nets.generate_identification_model(input_shape=(50, 50, 1), number_of_classes=2),91            src.nets.generate_identification_model(input_shape=(50, 50, 1), number_of_classes=4),92            src.nets.generate_identification_model(input_shape=(50, 50, 1), number_of_classes=2)93        ))9495        load_path_identification = load_folder + "identification.h5" if load_folder is not None else None9697        net_identification.train_models(number_of_models=6,98                                        epochs=args.identification_epochs,99                                        steps_per_epoch=args.steps_per_epoch,100                                        weights_save_path=save_folder + "identification.h5",101                                        weights_load_path=None,102                                        do_checkpoint=args.do_checkpoint,103                                        generators=IdentificationNetwork.generators(104                                            folder_images=datasets_folder + "Identification/",105                                            batch_size=args.batch_size,106                                            image_ext=(".jpg", ".png")
...

classifier.py

Source:classifier.py

1import torch2import torch.nn as nn3from torch.utils.checkpoint import checkpoint4from tortoise.models.arch_util import Upsample, Downsample, normalization, zero_module, AttentionBlock5class ResBlock(nn.Module):6    def __init__(7        self,8        channels,9        dropout,10        out_channels=None,11        use_conv=False,12        use_scale_shift_norm=False,13        dims=2,14        up=False,15        down=False,16        kernel_size=3,17        do_checkpoint=True,18    ):19        super().__init__()20        self.channels = channels21        self.dropout = dropout22        self.out_channels = out_channels or channels23        self.use_conv = use_conv24        self.use_scale_shift_norm = use_scale_shift_norm25        self.do_checkpoint = do_checkpoint26        padding = 1 if kernel_size == 3 else 227        self.in_layers = nn.Sequential(28            normalization(channels),29            nn.SiLU(),30            nn.Conv1d(channels, self.out_channels, kernel_size, padding=padding),31        )32        self.updown = up or down33        if up:34            self.h_upd = Upsample(channels, False, dims)35            self.x_upd = Upsample(channels, False, dims)36        elif down:37            self.h_upd = Downsample(channels, False, dims)38            self.x_upd = Downsample(channels, False, dims)39        else:40            self.h_upd = self.x_upd = nn.Identity()41        self.out_layers = nn.Sequential(42            normalization(self.out_channels),43            nn.SiLU(),44            nn.Dropout(p=dropout),45            zero_module(46                nn.Conv1d(self.out_channels, self.out_channels, kernel_size, padding=padding)47            ),48        )49        if self.out_channels == channels:50            self.skip_connection = nn.Identity()51        elif use_conv:52            self.skip_connection = nn.Conv1d(53                dims, channels, self.out_channels, kernel_size, padding=padding54            )55        else:56            self.skip_connection = nn.Conv1d(dims, channels, self.out_channels, 1)57    def forward(self, x):58        if self.do_checkpoint:59            return checkpoint(60                self._forward, x61            )62        else:63            return self._forward(x)64    def _forward(self, x):65        if self.updown:66            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]67            h = in_rest(x)68            h = self.h_upd(h)69            x = self.x_upd(x)70            h = in_conv(h)71        else:72            h = self.in_layers(x)73        h = self.out_layers(h)74        return self.skip_connection(x) + h75class AudioMiniEncoder(nn.Module):76    def __init__(self,77                 spec_dim,78                 embedding_dim,79                 base_channels=128,80                 depth=2,81                 resnet_blocks=2,82                 attn_blocks=4,83                 num_attn_heads=4,84                 dropout=0,85                 downsample_factor=2,86                 kernel_size=3):87        super().__init__()88        self.init = nn.Sequential(89            nn.Conv1d(spec_dim, base_channels, 3, padding=1)90        )91        ch = base_channels92        res = []93        self.layers = depth94        for l in range(depth):95            for r in range(resnet_blocks):96                res.append(ResBlock(ch, dropout, do_checkpoint=False, kernel_size=kernel_size))97            res.append(Downsample(ch, use_conv=True, out_channels=ch*2, factor=downsample_factor))98            ch *= 299        self.res = nn.Sequential(*res)100        self.final = nn.Sequential(101            normalization(ch),102            nn.SiLU(),103            nn.Conv1d(ch, embedding_dim, 1)104        )105        attn = []106        for a in range(attn_blocks):107            attn.append(AttentionBlock(embedding_dim, num_attn_heads, do_checkpoint=False))108        self.attn = nn.Sequential(*attn)109        self.dim = embedding_dim110    def forward(self, x):111        h = self.init(x)112        h = self.res(h)113        h = self.final(h)114        for blk in self.attn:115            h = checkpoint(blk, h)116        return h[:, :, 0]117class AudioMiniEncoderWithClassifierHead(nn.Module):118    def __init__(self, classes, distribute_zero_label=True, **kwargs):119        super().__init__()120        self.enc = AudioMiniEncoder(**kwargs)121        self.head = nn.Linear(self.enc.dim, classes)122        self.num_classes = classes123        self.distribute_zero_label = distribute_zero_label124    def forward(self, x, labels=None):125        h = self.enc(x)126        logits = self.head(h)127        if labels is None:128            return logits129        else:130            if self.distribute_zero_label:131                oh_labels = nn.functional.one_hot(labels, num_classes=self.num_classes)132                zeros_indices = (labels == 0).unsqueeze(-1)133                # Distribute 20% of the probability mass on all classes when zero is specified, to compensate for dataset noise.134                zero_extra_mass = torch.full_like(oh_labels, dtype=torch.float, fill_value=.2/(self.num_classes-1))135                zero_extra_mass[:, 0] = -.2136                zero_extra_mass = zero_extra_mass * zeros_indices137                oh_labels = oh_labels + zero_extra_mass138            else:139                oh_labels = labels140            loss = nn.functional.cross_entropy(logits, oh_labels)...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.