lpd 0.1.3

A Fast, Flexible Trainer with Callbacks and Extensions for PyTorch

lpd

A Fast, Flexible Trainer with Callbacks and Extensions for PyTorch

lpd derives from the Hebrew word lapid (לפיד) which means "torch".

For latest PyPI stable release

    pip install lpd

Usage

lpd intended to properly structure your pytorch model training. The main usages are given below.

Training your model

    from lpd.trainer import Trainer
    import lpd.utils.torch_utils as tu
    import lpd.utils.general_utils as gu
    from lpd.enums impoCallbackPhase, TrainerState, MonitorType, MonitorMode, StatsType
    from lpd.callbacks import StatsPrint, ModelCheckPoint, Tensorboard, EarlyStopping, SchedulerStep
    from lpd.extensions.custom_metrics import binary_accuracy_with_logits

    gu.seed_all(seed=42)

    device = tu.get_gpu_device_if_available() # with fallback to CPU if GPU not avilable
    model = TestModel(config, num_embeddings).to(device) #this is your model class, and its being sent to the relevant device
    optimizer = optim.SGD(params=model.parameters())
    scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', patience=5, verbose=True)
    loss_func = nn.BCEWithLogitsLoss().to(device) #this is your loss class, already sent to the relevant device
    metric_name_to_func = {'acc':binary_accuracy_with_logits} # add as much metrics as you like

    # you can use some of the defined callbacks, or you can create your own
    callbacks = [
                SchedulerStep(scheduler_parameters_func=lambda trainer: trainer.val_stats.get_loss()), # notice lambda for scheduler that takes loss in step()
                ModelCheckPoint(checkpoint_dir, checkpoint_file_name, MonitorType.LOSS, StatsType.VAL, MonitorMode.MIN, save_best_only=True), 
                Tensorboard(summary_writer_dir=summary_writer_dir),
                EarlyStopping(patience=10, MonitorType.METRIC, StatsType.VAL, MonitorMode.MAX, metric_name='acc'),
                StatsPrint(metric_names=metric_name_to_func.keys())
            ]

    trainer = Trainer(model, 
                      device, 
                      loss_func, 
                      optimizer,
                      scheduler,
                      metric_name_to_func, 
                      train_data_loader,  # DataLoader, Iterable or Generator
                      val_data_loader,    # DataLoader, Iterable or Generator
                      train_steps,
                      val_steps,
                      num_epochs,
                      callbacks)
    
    trainer.train()

Evaluating your model

    trainer.evaluate(test_data_loader, test_steps)

TrainerStats

Trainer tracks stats for train/val/test and you can access them in your custom callbacks or any other place you see fit.

Here are some examples

    train_loss = trainer.train_stats.get_loss()         # the mean of the last epoch's train losses
    val_loss = trainer.val_stats.get_loss()             # the mean of the last epoch's val losses

    train_metrics = trainer.train_stats.get_metrics()   # dictionary metric_name->mean of the last epoch's train metrics
    val_metrics = trainer.val_stats.get_metrics()       # dictionary metric_name->mean of the last epoch's val metrics

Callbacks

Some common callbacks are available under lpd.callbacks.

Notice that cb_phase (lpd.enums.CallbackPhase) will determine the execution phase,

and that apply_on_states (lpd.enums.TrainerState) will determine the execution state

These are the current available phases and states, more might be added in future releases

        TrainerState.EXTERNAL
        CallbackPhase.TRAIN_BEGIN
        # train loop:
            CallbackPhase.EPOCH_BEGIN

            TrainerState.TRAIN
            # batches loop:
                CallbackPhase.BATCH_BEGIN
                # batch
                CallbackPhase.BATCH_END
            TrainerState.VAL
            # batches loop:
                CallbackPhase.BATCH_BEGIN
                # batch
                CallbackPhase.BATCH_END
            TrainerState.EXTERNAL

            CallbackPhase.EPOCH_END
        CallbackPhase.TRAIN_END

Evaluation phases and states will be behave as follow

        TrainerState.EXTERNAL
        CallbackPhase.TEST_BEGIN
        TrainerState.TEST
        # batches loop:
            CallbackPhase.BATCH_BEGIN
            # batch
            CallbackPhase.BATCH_END
        TrainerState.EXTERNAL
        CallbackPhase.TEST_END

With phases and states you'll have full control over the timing of your callbacks,

so for example, say you need SchedulerStep callback to control your scheduler,

but only at the end of every batch, and only when in train state (as oppose to validation and test) then define your SchedulerStep callback like so:

    from lpd.callbacks import SchedulerStep
    from lpd.enums import CallbackPhase, TrainerState
    SchedulerStep(cb_phase=CallbackPhase.BATCH_END, apply_on_states=TrainerState.TRAIN)

In case you need it on validation state as well, pass a list for apply_on_states like so:

    SchedulerStep(cb_phase=CallbackPhase.BATCH_END, apply_on_states=[TrainerState.TRAIN, TrainerState.VAL])

Below is an output example for StatsPrint callback that will print an epoch summary at the end of every epoch

You can also create your own custom callbacks

    import lpd.enums as en
    from lpd.callbacks import CallbackBase

    class MyAwesomeCallback(CallbackBase):
        def __init__(self, cb_phase=CallbackPhase.BATCH_END, apply_on_states=[TrainerState.TRAIN, TrainerState.VAL]):
            super(MyAwesomeCallback, self).__init__(cb_phase, apply_on_states)

        def __call__(self, callback_context): # <=== implement this method!
            # your implementation here
            # using callback_context, you can access anything in your trainer
            # below are some examples to get the hang of it
            val_loss = callback_context.val_stats.get_loss()
            train_loss = callback_context.train_stats.get_loss()
            train_metrics = callback_context.train_stats.get_metrics()
            val_metrics = callback_context.val_stats.get_metrics()
            opt = callback_context.trainer.optimizer
            scheduler = callback_context.trainer.scheduler

            # you can also mark the trainer as STOP by calling the stop_training() method
            if val_loss < 0.0001:
                callback_context.trainer.stop_training()

Lets expand MyAwesomeCallback with CallbackMonitor to track if our validation loss is getting better

    from lpd.callbacks import CallbackBase, CallbackMonitor # <== CallbackMonitor added
    from lpd.enums import CallbackPhase, TrainerState, MonitorType, StatsType, MonitorMode # <== added few needed enums to configure CallbackMonitor

    class MyAwesomeCallback(CallbackBase):
        def __init__(self, cb_phase=CallbackPhase.BATCH_END, apply_on_states=[TrainerState.TRAIN, TrainerState.VAL]):
            super(MyAwesomeCallback, self).__init__(cb_phase, apply_on_states)
            
            # adding CallbackMonitor to track VAL LOSS with regards to MIN (lower is better)
            self.val_loss_monitor = CallbackMonitor(patience=20, MonitorType.LOSS, StatsType.VAL, MonitorMode.MIN)

        def __call__(self, callback_context: CallbackContext): # <=== implement this method!
            # same as before, using callback_context, you can access anything in your trainer
            train_metrics = callback_context.train_stats.get_metrics()
            val_metrics = callback_context.val_stats.get_metrics()

            # invoke track() method with callback_context
            monitor_result = self.val_loss_monitor.track(callback_context)

            # monitor_result (lpd.callbacks.CallbackMonitorResult) contains lots of informative properties
            # for example, lets check the status of the patience countdown

            if monitor_result.has_patience():
                print(f'[MyAwesomeCallback] - patience count: {monitor_result.patience_left}')

            # Or, lets stop the trainer (by calling the trainer.stop_training() ) 
            # if our monitored value did not improve

            if not monitor_result.has_improved():
                print(f'[MyAwesomeCallback] - {monitor_result.description} has stopped improving')
                callback_context.trainer.stop_training()

Utils

lpd.utils provides few utils files (torch_utils, file_utils and general_utils) For example, a good practice is to use

    import lpd.utils.general_utils as gu
    gu.seed_all(seed=42)  # because its the answer to life and the universe

As early as possible in your code, to make sure that results are reproducible

Extensions

lpd.extensions provides some custom pytorch layers, these are just some layers we like using when we create our models, to gain better flexibility.

So you can use them at your own will, there youll also find custom metrics and schedulers. We will add more layers, metrics and schedulers from time to time.

TODOS (more added frequently)

• Add callback descriptions to summary
• Add support for multiple schedulers
• Add support for multiple losses
• Save trainer in checkpoint to enable loading a model and continue training from last checkpoint
• Add colab examples

Something is missing?! please share with us

You can open an issue, but also feel free to email us at [e-mail]