hive.utils.loggers module

class hive.utils.loggers.Logger(timescales=None)

Bases: abc.ABC, hive.utils.registry.Registrable

Abstract class for logging in hive.

Constructor for base Logger class. Every Logger must call this constructor in its own constructor

Parameters

timescales (str | list(str)) – The different timescales at which logger needs to log. If only logging at one timescale, it is acceptable to only pass a string.

register_timescale(timescale)

Register a new timescale with the logger.

Parameters

timescale (str) – Timescale to register.

abstract log_config(config)

Log the config.

Parameters

config (dict) – Config parameters.

abstract log_scalar(name, value, prefix)

Log a scalar variable.

Parameters

• name (str) – Name of the metric to be logged.

• value (float) – Value to be logged.

• prefix (str) – Prefix to append to metric name.

abstract log_metrics(metrics, prefix)

Log a dictionary of values.

Parameters

• metrics (dict) – Dictionary of metrics to be logged.

• prefix (str) – Prefix to append to metric name.

abstract save(dir_name)

Saves the current state of the log files.

Parameters

dir_name (str) – Name of the directory to save the log files.

abstract load(dir_name)

Loads the log files from given directory.

Parameters

dir_name (str) – Name of the directory to load the log file from.

classmethod type_name()

This should represent a string that denotes the which type of class you are creating. For example, “logger”, “agent”, or “env”.

class hive.utils.loggers.ScheduledLogger(timescales=None, logger_schedules=None)

Bases: hive.utils.loggers.Logger

Abstract class that manages a schedule for logging.

The update_step method should be called for each step in the loop to update the logger’s schedule. The should_log method can be used to check whether the logger should log anything.

This schedule is not strictly enforced! It is still possible to log something even if should_log returns false. These functions are just for the purpose of convenience.

Any timescales not assigned schedule from logger_schedules will be assigned a ConstantSchedule(True).

Parameters

• timescales (str|list[str]) – The different timescales at which logger needs to log. If only logging at one timescale, it is acceptable to only pass a string.

• logger_schedules (Schedule|list|dict) – Schedules used to keep track of when to log. If a single schedule, it is copied for each timescale. If a list of schedules, the schedules are matched up in order with the list of timescales provided. If a dictionary, the keys should be the timescale and the values should be the schedule.

register_timescale(timescale, schedule=None)

Register a new timescale.

Parameters

• timescale (str) – Timescale to register.

• schedule (Schedule) – Schedule to use for this timescale.

update_step(timescale)

Update the step and schedule for a given timescale.

Parameters

timescale (str) – A registered timescale.

should_log(timescale)

Check if you should log for a given timescale.

Parameters

timescale (str) – A registered timescale.

save(dir_name)

Saves the current state of the log files.

Parameters

dir_name (str) – Name of the directory to save the log files.

load(dir_name)

Loads the log files from given directory.

Parameters

dir_name (str) – Name of the directory to load the log file from.

class hive.utils.loggers.NullLogger(timescales=None, logger_schedules=None)

Bases: hive.utils.loggers.ScheduledLogger

A null logger that does not log anything.

Used if you don’t want to log anything, but still want to use parts of the framework that ask for a logger.

Any timescales not assigned schedule from logger_schedules will be assigned a ConstantSchedule(True).

Parameters

• timescales (str|list[str]) – The different timescales at which logger needs to log. If only logging at one timescale, it is acceptable to only pass a string.

• logger_schedules (Schedule|list|dict) – Schedules used to keep track of when to log. If a single schedule, it is copied for each timescale. If a list of schedules, the schedules are matched up in order with the list of timescales provided. If a dictionary, the keys should be the timescale and the values should be the schedule.

log_config(config)

Log the config.

Parameters

config (dict) – Config parameters.

log_scalar(name, value, timescale)

Log a scalar variable.

Parameters

• name (str) – Name of the metric to be logged.

• value (float) – Value to be logged.

• prefix (str) – Prefix to append to metric name.

log_metrics(metrics, timescale)

Log a dictionary of values.

Parameters

• metrics (dict) – Dictionary of metrics to be logged.

• prefix (str) – Prefix to append to metric name.

save(dir_name)

Saves the current state of the log files.

Parameters

dir_name (str) – Name of the directory to save the log files.

load(dir_name)

Loads the log files from given directory.

Parameters

dir_name (str) – Name of the directory to load the log file from.

class hive.utils.loggers.WandbLogger(timescales=None, logger_schedules=None, project=None, name=None, dir=None, mode=None, id=None, resume=None, start_method=None, **kwargs)

Bases: hive.utils.loggers.ScheduledLogger

A Wandb logger.

This logger can be used to log to wandb. It assumes that wandb is configured locally on your system. Multiple timescales/loggers can be implemented by instantiating multiple loggers with different logger_names. These should still have the same project and run names.

Check the wandb documentation for more details on the parameters.

Parameters

• timescales (str|list[str]) – The different timescales at which logger needs to log. If only logging at one timescale, it is acceptable to only pass a string.

• logger_schedules (Schedule|list|dict) – Schedules used to keep track of when to log. If a single schedule, it is copied for each timescale. If a list of schedules, the schedules are matched up in order with the list of timescales provided. If a dictionary, the keys should be the timescale and the values should be the schedule.

• project (str) – Name of the project. Wandb’s dash groups all runs with the same project name together.

• name (str) – Name of the run. Used to identify the run on the wandb dash.

• dir (str) – Local directory where wandb saves logs.

• mode (str) – The mode of logging. Can be “online”, “offline” or “disabled”. In offline mode, writes all data to disk for later syncing to a server, while in disabled mode, it makes all calls to wandb api’s noop’s, while maintaining core functionality.

• id (str, optional) – A unique ID for this run, used for resuming. It must be unique in the project, and if you delete a run you can’t reuse the ID.

• resume (bool, str, optional) – Sets the resuming behavior. Options are the same as mentioned in Wandb’s doc.

• start_method (str) – The start method to use for wandb’s process. See https://docs.wandb.ai/guides/track/launch#init-start-error.

• **kwargs – You can pass any other arguments to wandb’s init method as keyword arguments. Note, these arguments can’t be overriden from the command line.

log_config(config)

Log the config.

Parameters

config (dict) – Config parameters.

log_scalar(name, value, prefix)

Log a scalar variable.

Parameters

• name (str) – Name of the metric to be logged.

• value (float) – Value to be logged.

• prefix (str) – Prefix to append to metric name.

log_metrics(metrics, prefix)

Log a dictionary of values.

Parameters

• metrics (dict) – Dictionary of metrics to be logged.

• prefix (str) – Prefix to append to metric name.

class hive.utils.loggers.ChompLogger(timescales=None, logger_schedules=None)

Bases: hive.utils.loggers.ScheduledLogger

This logger uses the Chomp data structure to store all logged values which are then directly saved to disk.

Any timescales not assigned schedule from logger_schedules will be assigned a ConstantSchedule(True).

Parameters

• timescales (str|list[str]) – The different timescales at which logger needs to log. If only logging at one timescale, it is acceptable to only pass a string.

• logger_schedules (Schedule|list|dict) – Schedules used to keep track of when to log. If a single schedule, it is copied for each timescale. If a list of schedules, the schedules are matched up in order with the list of timescales provided. If a dictionary, the keys should be the timescale and the values should be the schedule.

log_config(config)

Log the config.

Parameters

config (dict) – Config parameters.

log_scalar(name, value, prefix)

Log a scalar variable.

Parameters

• name (str) – Name of the metric to be logged.

• value (float) – Value to be logged.

• prefix (str) – Prefix to append to metric name.

log_metrics(metrics, prefix)

Log a dictionary of values.

Parameters

• metrics (dict) – Dictionary of metrics to be logged.

• prefix (str) – Prefix to append to metric name.

save(dir_name)

Saves the current state of the log files.

Parameters

dir_name (str) – Name of the directory to save the log files.

load(dir_name)

Loads the log files from given directory.

Parameters

dir_name (str) – Name of the directory to load the log file from.

class hive.utils.loggers.CompositeLogger(logger_list)

Bases: hive.utils.loggers.Logger

This Logger aggregates multiple loggers together.

This logger is for convenience and allows for logging using multiple loggers without having to keep track of several loggers. When timescales are updated, this logger updates the timescale for each one of its component loggers. When logging, logs to each of its component loggers as long as the logger is not a ScheduledLogger that should not be logging for the timescale.

Constructor for base Logger class. Every Logger must call this constructor in its own constructor

Parameters

timescales (str | list(str)) – The different timescales at which logger needs to log. If only logging at one timescale, it is acceptable to only pass a string.

register_timescale(timescale, schedule=None)

Register a new timescale with the logger.

Parameters

timescale (str) – Timescale to register.

log_config(config)

Log the config.

Parameters

config (dict) – Config parameters.

log_scalar(name, value, prefix)

Log a scalar variable.

Parameters

• name (str) – Name of the metric to be logged.

• value (float) – Value to be logged.

• prefix (str) – Prefix to append to metric name.

log_metrics(metrics, prefix)

Log a dictionary of values.

Parameters

• metrics (dict) – Dictionary of metrics to be logged.

• prefix (str) – Prefix to append to metric name.

update_step(timescale)

Update the step and schedule for a given timescale for every ScheduledLogger.

Parameters

timescale (str) – A registered timescale.

should_log(timescale)

Check if you should log for a given timescale. If any logger in the list is scheduled to log, returns True.

Parameters

timescale (str) – A registered timescale.

save(dir_name)

Saves the current state of the log files.

Parameters

dir_name (str) – Name of the directory to save the log files.

load(dir_name)

Loads the log files from given directory.

Parameters

dir_name (str) – Name of the directory to load the log file from.