minrecord.manager¶

minrecord.manager ¶

Contain a record manager implementation.

minrecord.manager.RecordManager ¶

Implement a simple record manager.

This class proposes an approach to manage a group of records, but it is possible to use other approaches. If this class does not fit your needs, feel free to use another approach.

Parameters:

Name	Type	Description	Default
`records`	`dict[str, BaseRecord[Any]] \| None`	The initial records to add to the manager.	`None`

Example

>>> from minrecord import RecordManager, MinScalarRecord
>>> manager = RecordManager()
>>> manager.add_record(MinScalarRecord("loss"))
>>> manager.get_record("loss")
MinScalarRecord(name=loss, max_size=10, size=0)
>>> manager.get_record("new_record")
Record(name=new_record, max_size=10, size=0)

minrecord.manager.RecordManager.add_record ¶

add_record(
    record: BaseRecord[Any],
    key: str | None = None,
    exist_ok: bool = False,
) -> None

Add a record to the manager.

Parameters:

Name	Type	Description	Default
`record`	`BaseRecord[Any]`	The record to add to the manager.	required
`key`	`str \| None`	The key to store the record. If `None`, the name of the record is used.	`None`
`exist_ok`	`bool`	If `False`, `RuntimeError` is raised if the key already exists. This parameter should be set to `True` to overwrite the record for this key.	`False`

Raises:

Type	Description
`RuntimeError`	if a record is already registered for the key and `exist_ok=False`.

Example

>>> from minrecord import RecordManager, MinScalarRecord
>>> manager = RecordManager()
>>> manager.add_record(MinScalarRecord("loss"))
>>> manager
RecordManager(
  (loss): MinScalarRecord(name=loss, max_size=10, size=0)
)
>>> manager.add_record(MinScalarRecord("loss"), "my key")
>>> manager
RecordManager(
  (loss): MinScalarRecord(name=loss, max_size=10, size=0)
  (my key): MinScalarRecord(name=loss, max_size=10, size=0)
)

minrecord.manager.RecordManager.get_best_values ¶

get_best_values(
    prefix: str = "", suffix: str = ""
) -> dict[str, Any]

Get the best value of each metric.

This method ignores the metrics with empty record and the non-comparable record.

Parameters:

Name	Type	Description	Default
`prefix`	`str`	The prefix used to create the dict of best values. The goal of this prefix is to generate a name which is different from the metric name to avoid confusion. By default, the returned dict uses the same name as the metric.	`''`
`suffix`	`str`	The suffix used to create the dict of best values. The goal of this suffix is to generate a name which is different from the metric name to avoid confusion. By default, the returned dict uses the same name as the metric.	`''`

Returns:

Type	Description
`dict[str, Any]`	The dict with the best value of each metric.

Example

>>> from minrecord import RecordManager, MaxScalarRecord
>>> manager = RecordManager()
>>> manager.add_record(MaxScalarRecord("accuracy"))
>>> manager.get_record("accuracy").add_value(42.0)
>>> manager.get_best_values()
{'accuracy': 42.0}
>>> manager.get_best_values(prefix="best/")
{'best/accuracy': 42.0}
>>> manager.get_best_values(suffix="/best")
{'accuracy/best': 42.0}

minrecord.manager.RecordManager.get_record ¶

get_record(key: str) -> BaseRecord[Any]

Get the record associated to a key.

Parameters:

Name	Type	Description	Default
`key`	`str`	The key of the record to retrieve.	required

Returns:

Type	Description
`BaseRecord[Any]`	The record if it exists, otherwise it returns an empty record. The created empty record is a `Record` object.

Example

>>> from minrecord import RecordManager, MinScalarRecord
>>> manager = RecordManager()
>>> manager.add_record(MinScalarRecord("loss"))
>>> manager.get_record("loss")
MinScalarRecord(name=loss, max_size=10, size=0)
>>> manager.get_record("new_record")
Record(name=new_record, max_size=10, size=0)

minrecord.manager.RecordManager.get_records ¶

get_records() -> dict[str, BaseRecord[Any]]

Get all the records.

Returns:

Type	Description
`dict[str, BaseRecord[Any]]`	The records with their associated keys.

Example

>>> from minrecord import RecordManager, MinScalarRecord
>>> manager = RecordManager()
>>> manager.add_record(MinScalarRecord("loss"))
>>> manager.get_records()
{'loss': MinScalarRecord(name=loss, max_size=10, size=0)}

minrecord.manager.RecordManager.has_record ¶

has_record(key: str) -> bool

Indicate if the engine has a record for the given key.

Parameters:

Name	Type	Description	Default
`key`	`str`	The key of the record to check.	required

Returns:

Type	Description
`bool`	`True` if the record exists, `False` otherwise

Example

>>> from minrecord import RecordManager, MinScalarRecord
>>> manager = RecordManager()
>>> manager.add_record(MinScalarRecord("loss"))
>>> manager.has_record("loss")
True
>>> manager.has_record("missing")
False

minrecord.manager.RecordManager.load_state_dict ¶

load_state_dict(state_dict: dict[str, Any]) -> None

Load the state values from a dict.

Parameters:

Name	Type	Description	Default
`state_dict`	`dict[str, Any]`	A dict with the new state values.	required

Example

>>> from minrecord import RecordManager, Record
>>> manager = RecordManager()
>>> manager.add_record(Record("value"))
>>> manager.load_state_dict({"value": {"state": {"record": ((0, 1), (1, 0.5), (2, 0.25))}}})
>>> manager.get_record("value").get_last_value()
0.25

minrecord.manager.RecordManager.state_dict ¶

state_dict() -> dict[str, Any]

Return a dictionary containing state values of all the records.

Returns:

Type	Description
`dict[str, Any]`	The dictionary containing state values of all the records.

Example

>>> from minrecord import RecordManager
>>> manager = RecordManager()
>>> manager.state_dict()
{}