context: Context object

In luz: Higher Level 'API' for 'torch'

Context object

Description

Context object storing information about the model training context. See also ctx.

Public fields

buffers

This is a list of buffers that callbacks can use to write temporary information into ctx.

Active bindings

records

stores information about values logged with self$log.

device

allows querying the current accelerator device

callbacks

list of callbacks that will be called.

iter

current iteration

batch

the current batch data. a list with input data and targets.

input

a shortcut for ctx$batch[[1]]

target

a shortcut for ctx$batch[[2]]

min_epochs

the minimum number of epochs that the model will run on.

max_epochs

the maximum number of epochs that the model will run.

hparams

a list of hyperparameters that were used to initialize ctx$model.

opt_hparams

a list of hyperparameters used to initialize the ctx$optimizers.

train_data

a dataloader that is used for training the model

valid_data

a dataloader using during model validation

accelerator

an accelerator() used to move data, model and etc the the correct device.

optimizers

a named list of optimizers that will be used during model training.

verbose

bool wether the process is in verbose mode or not.

handlers

List of error handlers that can be used. See rlang::try_fetch() for more info.

epoch_handlers

List of error handlers that can be used. See rlang::try_fetch() for more info.

training

A bool indicating if the model is in training or validation mode.

model

The model being trained.

pred

Last predicted values.

opt

Current optimizer.

opt_name

Current optimizer name.

data

Current dataloader in use.

loss_fn

Loss function used to train the model

loss

Last computed loss values. Detached from the graph.

loss_grad

Last computed loss value, not detached, so you can do additional tranformation.

epoch

Current epoch.

metrics

List of metrics that are tracked by the process.

Methods

Public methods

• context$new()

• context$log()

• context$log_metric()

• context$get_log()

• context$get_metrics()

• context$get_metric()

• context$get_formatted_metrics()

• context$get_metrics_df()

• context$set_verbose()

• context$clean()

• context$call_callbacks()

• context$state_dict()

• context$unsafe_set_records()

• context$clone()

---

Method new()

Initializes the context object with minimal necessary information.

Usage

context$new(verbose, accelerator, callbacks, training)

Arguments

verbose

Whether the context should be in verbose mode or not.

accelerator

A luz accelerator() that configures device placement and others.

callbacks

A list of callbacks used by the model. See luz_callback().

training

A boolean that indicates if the context is in training mode or not.

---

Method log()

Allows logging arbitrary information in the ctx.

Usage

context$log(what, set, value, index = NULL, append = TRUE)

Arguments

what

(string) What you are logging.

set

(string) Usually 'train' or 'valid' indicating the set you want to lot to. But can be arbitrary info.

value

value to log

value

Arbitrary value to log.

index

Index that this value should be logged. If NULL the value is added to the end of list, otherwise the index is used.

append

If TRUE and a value in the corresponding index already exists, then value is appended to the current value. If FALSE value is overwritten in favor of the new value.

---

Method log_metric()

Log a metric gen its name and value. Metric values are indexed by epoch.

Usage

context$log_metric(name, value)

Arguments

name

name of the metric

value

value to log

value

Arbitrary value to log.

---

Method get_log()

Get a specific value from the log.

Usage

context$get_log(what, set, index = NULL)

Arguments

what

(string) What you are logging.

set

(string) Usually 'train' or 'valid' indicating the set you want to lot to. But can be arbitrary info.

index

Index that this value should be logged. If NULL the value is added to the end of list, otherwise the index is used.

---

Method get_metrics()

Get all metric given an epoch and set.

Usage

context$get_metrics(set, epoch = NULL)

Arguments

set

(string) Usually 'train' or 'valid' indicating the set you want to lot to. But can be arbitrary info.

epoch

The epoch you want to extract metrics from.

---

Method get_metric()

Get the value of a metric given its name, epoch and set.

Usage

context$get_metric(name, set, epoch = NULL)

Arguments

name

name of the metric

set

(string) Usually 'train' or 'valid' indicating the set you want to lot to. But can be arbitrary info.

epoch

The epoch you want to extract metrics from.

---

Method get_formatted_metrics()

Get formatted metrics values

Usage

context$get_formatted_metrics(set, epoch = NULL)

Arguments

set

(string) Usually 'train' or 'valid' indicating the set you want to lot to. But can be arbitrary info.

epoch

The epoch you want to extract metrics from.

---

Method get_metrics_df()

Get a data.frame containing all metrics.

Usage

context$get_metrics_df()

---

Method set_verbose()

Allows setting the verbose attribute.

Usage

context$set_verbose(verbose = NULL)

Arguments

verbose

boolean. If TRUE verbose mode is used. If FALSE non verbose. if NULL we use the result of interactive().

---

Method clean()

Removes unnecessary information from the context object.

Usage

context$clean()

---

Method call_callbacks()

Call the selected callbacks. Where name is the callback types to call, eg 'on_epoch_begin'.

Usage

context$call_callbacks(name)

Arguments

name

name of the metric

---

Method state_dict()

Returns a list containing minimal information from the context. Used to create the returned values.

Usage

context$state_dict()

---

Method unsafe_set_records()

Are you sure you know what you are doing?

Usage

context$unsafe_set_records(records)

Arguments

records

New set of records to be set.

---

Method clone()

The objects of this class are cloneable with this method.

Usage

context$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

luz documentation built on April 17, 2023, 5:08 p.m.