mbapy.dl_torch.utils

This module provides functions and classes related to model initialization, training, and checkpoint management.

launch_visdom -> (visdom.Visdom, list)

General description

This function initializes the visdom server and returns the visdom object and an empty list for recording.

Returns

vis (visdom.Visdom): The initialized visdom object.
viz_record (list): An empty list for recording.

Notes

This function should be called before using any visualization functions.

Example

vis, viz_record = launch_visdom()

Mprint

General description

This class provides logging tools for printing and saving logs to a file.

Attrs

path (str): The path to the log file.
mode (str): The mode for logging, either "lazy" or "normal".
top_string (str): The top string for the log.
string (str): The log string.

Methods

mprint(*args) -> None: Prints the log and saves it to the log file based on the mode.
log_only(*args) -> None: Logs the message to the log file without printing.
exit(mode='a+') -> None: Writes the log string to the log file and closes it.
__str__() -> str: Returns the string representation of the Mprint object.

Notes

The mprint method prints the log and saves it to the log file based on the mode.
The log_only method logs the message to the log file without printing.
The exit method writes the log string to the log file and closes it.

Example

logger = Mprint("log.txt", "normal")
logger.mprint("Logging message")
logger.exit()

GlobalSettings

General description

This class contains global settings and hyperparameters for the model.

Attrs

read (dict): Dictionary for data reading.
batch_size (int): Batch size for data loading.
load_shape (list): Shape for data loading.
model (None): The model object.
arch (None): The architecture of the model.
lr (float): Learning rate.
in_shape (list): Input shape for the model.
out_shape (list): Output shape for the model.
load_db_ratio (float): Load database ratio.
epochs (int): Total number of epochs.
now_epoch (int): Current epoch number.
left_epochs (int): Number of epochs left.
print_freq (int): Frequency of printing logs.
test_freq (int): Frequency of testing the model.
momentum (float): Momentum for optimization.
weight_decay (float): Weight decay for optimization.
seed (int): Random seed for reproducibility.
moco_m (float): Moco m parameter.
moco_k (int): Moco k parameter.
byolq_k (int): Byolq k parameter.
moco_t (float): Moco t parameter.
cos (bool): Cosine flag.
paths (dict): Dictionary for paths.
data (str): Data information.
model_root (str): Root directory for the model.
resume_paths (list): List of paths for resuming the model.
resume (str): Path for resuming the model.
mp (Mprint): Mprint object for logging.

Methods

add_epoch(addon: int) -> bool: Adds an epoch and returns True if the current epoch is greater than the total epochs.
to_dict(printOut=False, mp=None) -> dict: Converts the object attributes to a dictionary and prints it if printOut is True.
set_resume() -> None: Sets the resume path for the model.

Notes

The add_epoch method adds an epoch and returns True if the current epoch is greater than the total epochs.
The to_dict method converts the object attributes to a dictionary and prints it if printOut is True.
The set_resume method sets the resume path for the model.

Example

mp = Mprint()
settings = GlobalSettings(mp, "model_root", seed=777)
settings.add_epoch()
settings.to_dict(printOut=True, mp=mp)
settings.set_resume()

init_model_parameter

General description

This function initializes the model parameters.

Params

model (torch.nn.Module): The model to be initialized.

Returns

model (torch.nn.Module): The initialized model.

Notes

This function initializes the model parameters using specific initialization methods for different types of layers.

Example

model = init_model_parameter(model)

adjust_learning_rate

General description

This function adjusts the learning rate of the given optimizer based on the current epoch and arguments.

Params

optimizer (torch.optim.Optimizer): Optimizer to adjust the learning rate of.
now_epoch (int): Current epoch number.
args (argparse.Namespace): Parsed command-line arguments.

Returns

None

Notes

This function adjusts the learning rate based on the cosine or stepwise learning rate schedule.

Example

adjust_learning_rate(optimizer, now_epoch, args)

format_secs

General description

This function formats a given number of seconds into hours, minutes, and seconds.

Params

sumSecs (int): Total number of seconds.

Returns

tuple: A tuple containing three integers representing the number of hours, minutes, and seconds respectively.

Notes

This function formats the total number of seconds into hours, minutes, and seconds.

Example

hours, minutes, seconds = format_secs(3600)

AverageMeter

General description

This class computes and stores the average and current value.

Attrs

name (str): Name of the meter.
fmt (str): Format for displaying the values.

Methods

reset() -> None: Resets the values of the meter.
update(val, n=1) -> None: Updates the meter with a new value and count.
__str__() -> str: Returns the string representation of the AverageMeter object.

Notes

This class computes and stores the average and current value for a specific metric.

ProgressMeter

General description

This class displays the progress of the training process.

Attrs

num_batches (int): Total number of batches.
meters (list): List of meters for displaying progress.
prefix (str): Prefix for the progress display.

Methods

display(batch) -> None: Displays the progress for the current batch.

Notes

This class displays the progress of the training process with the specified meters.

TimeLast

General description

This class calculates the time remaining for a given number of tasks.

Methods

update(left_tasks:int, just_done_tasks:int = 1) -> float: Updates the time remaining based on the number of tasks completed.

Notes

This class calculates the time remaining for a given number of tasks based on the time taken for the last task.

save_checkpoint

General description

This function saves a checkpoint of the model and optimizer state, along with other information.

Params

epoch (float): Current epoch number.
args (GlobalSettings): Global settings and hyperparameters.
model (torch.nn.Module): The model to be saved.
optimizer (torch.optim.Optimizer): The optimizer to be saved.
loss (float): Current loss value.
other (dict): Additional information to be saved.
tailName (str): A string for better identification in the checkpoint file name.

Notes

This function saves the model and optimizer state along with other information to a checkpoint file.

resume_checkpoint

General description

This function resumes the training from the last checkpoint if it exists, otherwise starts from scratch.

Params

args (GlobalSettings): Global settings and hyperparameters.
model (torch.nn.Module): Model to be trained.
optimizer (torch.optim.Optimizer): Optimizer to be used for training.

Returns

tuple: Tuple of the model, optimizer, and old_losses if checkpoint exists, otherwise tuple of model, optimizer, and 0.

Notes

This function resumes the training from the last checkpoint if it exists, otherwise starts from scratch.

viz_line

General description

This function visualizes a line plot using the visdom server.

Params

Y (float): Y-axis value.
X (float): X-axis value.
win (str): Window name for the plot.
title (str): Title for the plot.
name (str): Name for the line.
update (str): Update method for the plot.
opts (dict): Additional options for the plot.

Notes

This function visualizes a line plot using the visdom server and records the plot details.

re_viz_from_json_record

General description

This function re-visualizes the recorded plots from a JSON record file.

Params

path (str): Path to the JSON record file.

Notes

This function re-visualizes the recorded plots from a JSON record file.

Hope this helps! Let me know if you need anything else.