logger

experiment

class AugmentedSafeLoader(stream)

Bases: SafeLoader

Initialize the scanner.

construct_python_tuple(node)

Use PyYAML method for constructing a sequence to construct a tuple.

yaml_constructors = {'tag:yaml.org,2002:null': <function SafeConstructor.construct_yaml_null>, 'tag:yaml.org,2002:bool': <function SafeConstructor.construct_yaml_bool>, 'tag:yaml.org,2002:int': <function SafeConstructor.construct_yaml_int>, 'tag:yaml.org,2002:float': <function SafeConstructor.construct_yaml_float>, 'tag:yaml.org,2002:binary': <function SafeConstructor.construct_yaml_binary>, 'tag:yaml.org,2002:timestamp': <function SafeConstructor.construct_yaml_timestamp>, 'tag:yaml.org,2002:omap': <function SafeConstructor.construct_yaml_omap>, 'tag:yaml.org,2002:pairs': <function SafeConstructor.construct_yaml_pairs>, 'tag:yaml.org,2002:set': <function SafeConstructor.construct_yaml_set>, 'tag:yaml.org,2002:str': <function SafeConstructor.construct_yaml_str>, 'tag:yaml.org,2002:seq': <function SafeConstructor.construct_yaml_seq>, 'tag:yaml.org,2002:map': <function SafeConstructor.construct_yaml_map>, None: <function SafeConstructor.construct_undefined>, 'tag:yaml.org,2002:python/tuple': <function AugmentedSafeLoader.construct_python_tuple>}

class Experiment(env_name: str, algo_name: str, extra_info: Optional[str] = None, exp_id: Optional[str] = None, timestamp: Optional[datetime] = None, base_dir: str = '/home/user/SimuRLacra/Pyrado/pyrado/../data/temp', include_slurm_id: bool = True)

Bases: object

Class for defining experiments This is a path-like object, and as such it can be used everywhere a normal path would be used.

Experiment folder path: <base_dir>/<env_name>/<algo_name>/<timestamp>–<extra_info>

Constructor

Parameters:

• env_name – environment trained on

• algo_name – algorithm trained with, usually also includes the policy type, e.g. ‘a2c_fnn’

• extra_info – additional information on the experiment (free form)

• exp_id – combined timestamp and extra_info, usually the final folder name

• timestamp – experiment creation timestamp

• base_dir – base storage directory

• include_slurm_id – if a SLURM ID is present in the environment variables, include it in the experiment ID

matches(hint: str) → bool

Check if this experiment matches the given hint.

property prefix

Combination of experiment and algorithm

ask_for_experiment(latest_only: bool = False, max_display: int = 10, env_name: Optional[str] = None, temp: bool = True, perma: bool = True, hparam_list: Optional[List[str]] = None) → Experiment

Ask for an experiment on the console. This is the go-to entry point for evaluation scripts.

Parameters:

• latest_only – only select the latest experiment of each type (environment-algorithm combination)

• max_display – maximum number of items

• env_name – filter by env name

• temp – set to False to not look in the pyrado.TEMP directory

• perma – set to False to not look in the pyrado.PERMA directory

• hparam_list – load the hyper-parameter file and show the parameters in this list, sub-dicts can be separated with a dot

Returns:

query asking the user for an experiment

create_experiment_formatter(show_hparams: Optional[List[str]] = None, show_extra_info: bool = True) → Callable[[Experiment], str]

Returns an experiment formatter (i.e. a function that takes an experiment and produces a string) to be used in the ask-for-experiments dialog. It produces useful information like the timestamp based on the experiments’ data.

Parameters:

• show_hparams – list of “paths” to hyper-parameters that to be shown in the selection dialog; sub-dicts can be references with a dot, e.g. env.dt

• show_extra_info – whether to show the information stored in the extra_info field of the experiment

Returns:

a function that serves as the formatter

list_experiments(env_name: Optional[str] = None, algo_name: Optional[str] = None, base_dir: Optional[str] = None, *, temp: bool = True, perma: bool = True)

List all stored experiments.

Parameters:

• env_name – filter by env name

• algo_name – filter by algorithm name. Requires env_name to be used too

• base_dir – explicit base dir if desired. May also be a list of bases. Uses pyrado.TEMP_DIR and pyrado.EXP_DIR if not specified.

• temp – set to False to not look in the pyrado.TEMP directory

• perma – set to False to not look in the pyrado.PERMA directory

load_dict_from_yaml(yaml_file: str) → dict

Load a list of dicts (e.g. hyper-parameters) of an experiment from a YAML-file.

Parameters:

yaml_file – path to the YAML-file that

Returns:

a dict containing names as keys and a dict of parameter values

load_hyperparameters(ex_dir: PathLike, verbose: bool = True) → Optional[dict]

Loads the hyper-parameters-dict from the given experiment directory. The hyper-parameters file is assumed to be named hyperparams.yaml.

Parameters:

• ex_dir – experiment’s directory to load from

• verbose – if True, print message if no hyper-parameter file was found

save_dicts_to_yaml(*dicts: dict, save_dir: PathLike, file_name: str = 'hyperparams')

Save a list of dicts (e.g. hyper-parameters) of an experiment a YAML-file.

Parameters:

• dicts – dicts each containing a key (name) and a value (hyper-parameter)

• save_dir – directory to save the results in

• file_name – name of the YAML-file without suffix

select_by_hint(exps: Sequence[Experiment], hint: str)

Select experiment by hint.

setup_experiment(env_name: str, algo_name: str, extra_info: Optional[str] = None, base_dir: str = '/home/user/SimuRLacra/Pyrado/pyrado/../data/temp', include_slurm_id: bool = True)

Setup a new experiment for recording.

Parameters:

• env_name – environment trained on

• algo_name – algorithm trained with, usually also includes the policy type, e.g. ‘a2c_fnn’

• extra_info – additional information on the experiment (free form)

• base_dir – base storage directory

• include_slurm_id – if a SLURM ID is present in the environment variables, include them in the experiment ID

split_path_custom_common(path: ~typing.Union[str, ~pyrado.logger.experiment.Experiment]) -> (<class 'str'>, <class 'str'>)

Split a path at the point where the machine-dependent and the machine-independent part can be separated.

Parameters:

path – (complete) experiment path to be split

Returns:

name of the base directory (‘experiments’ for pyrado.EXP_DIR or ‘temp’ for pyrado.TEMP_DIR) where the experiment was located, and machine-independent part of the path

iteration

class IterationTracker

Bases: object

Track the current iteration/step number on multiple levels (for meta-algorithms)

Constructor

format(scope_sep='-', label_num_sep='_')

Format the current iteration stack into a string. Two parts can be customized:

Parameters:

• scope_sep – string separating the label and the number

• label_num_sep – string separating each label/number pair

Returns:

string with custom separators

get(label: str)

Get the iteration number for a labeled scope.

Parameters:

label – scope label

Returns:

iteration index

iteration(label: str, num: int)

Context with active iteration scope.

Parameters:

• label – scope label

• num – iteration index

peek() → tuple

pop() → tuple

Remove the last iteration scope.

push(label: str, num: int)

Push an iteration scope.

Parameters:

• label – scope label

• num – iteration index

step

class CSVPrinter(file: str)

Bases: StepLogPrinter

Logs step data to a CSV file

Constructor

Parameters:

file – csv file name

print_values(values: dict, ordered_keys: list, first_step: bool)

Print the values for a step.

Parameters:

• values – named progress values

• ordered_keys – value keys in a consistent order

• first_step – True for the first recorded step

class ConsolePrinter

Bases: StepLogPrinter

Prints step data to the console

print_values(values: dict, ordered_keys: list, first_step: bool)

Print the values for a step.

Parameters:

• values – named progress values

• ordered_keys – value keys in a consistent order

• first_step – True for the first recorded step

class LoggerAware

Bases: object

Base for objects holding a StepLogger. Features automatic detection of child LoggerAware objects. Override to customize.

property logger: StepLogger

Get or create the step logger to use for this object.

register_as_logger_parent(child)

class StepLogPrinter

Bases: ABC

Base class for log printers. Formats progress values for a step.

abstract print_values(values: dict, ordered_keys: list, first_step: bool)

Print the values for a step.

Parameters:

• values – named progress values

• ordered_keys – value keys in a consistent order

• first_step – True for the first recorded step

class StepLogger(print_intvl: int = 1)

Bases: object

Step-based progress logger. This class collects progress values during a step. At the end, the record_step function will pass the collected values to one or more StepLogPrinters. The logger also validates that no values are added unexpectedly, which i.e. a csv printer would not support.

Constructor

Parameters:

print_intvl – interval size, by default the logger records and prints on every call, i.e. every step

add_value(key: str, value, round_digits: Optional[int] = None)

Add a column value to the current step.

Parameters:

• key – data key

• value – value to record, pass ‘’ to print nothing

• round_digits – digits to rounds to, pass None (default) for no rounding

pop_prefix()

Remove the last string from the key prefix stack.

prefix(pfx: str)

Context manager to add a prefix to the key prefix stack during use.

Parameters:

pfx – prefix string

push_prefix(pfx)

Push a string onto the key prefix stack.

Parameters:

pfx – prefix string

record_step()

Record the currently stored values as step and print them at the end. To properly support nesting, this method does nothing if called twice in a row without an add_value in between.

class TensorBoardPrinter(dir)

Bases: StepLogPrinter

Class for writing tensorboard logs

Constructor

Parameters:

dir – folder path name

print_values(values: dict, ordered_keys: list, first_step: bool)

Print the values for a step.

Parameters:

• values – named progress values

• ordered_keys – value keys in a consistent order

• first_step – True for the first recorded step

create_csv_step_logger(save_dir: PathLike, file_name: str = 'progress.csv') → StepLogger

Create a step-based logger which only safes to a csv-file.

Parameters:

• save_dir – parent directory to save the results in (usually the algorithm’s save_dir)

• file_name – name of the cvs-file (with ending)

Returns:

step-based logger

Module contents

get_log_prefix_dir() → Optional[str]

Get the log prefix directory for this session.

resolve_log_path(path) → str

Resolve path against the global log dir if path is relative.

set_log_prefix_dir(log_dir: str)

Set the log prefix directory for this session.