3.10. Callbacks¶
Callbacks allow for further interaction with a running Optimization class. A useful callback could, for example, signal the optimization to stop after a certain time, or when overfitting.
Callback instances can be passed when a Optimization
instance is created:
callbacks = [Timeout(60*60), Logger()]
my_optimization = Optimization(*args, callbacks=callbacks)
3.10.1. Logger¶
-
class
Logger
(printfreq=100, names_to_log=None, path=None, writefreq_history=None, writefreq_bestparams=None, writefreq_datafiles=None, plot=False, kw_plot_history={}, kw_plot_contributions={}, kw_plot_residuals={})¶ Note
This callback is always included in an optimization, when running the ParAMS main script.
Combined callback that logs and saves the following data produced during an Optimization to disk for every Data Set provided:
- *history.dat: The (fx,x) value pair at every evaluation (Note: Unscaled x,
active
subset only). Plottable withparams plot
- *best_params: File or path of the parameter set x with the best f(x) so far
- residuals/ : Residuals of reference and predicted values \(y-\hat{y}\). Plottable with
params plot
- contributions/: Contributions of individual entries in the Data Set to the overall loss function value. Plottable with
params plot
Note
Including this callback with at least the default settings is generally recommended: without it, there will be no relevant data written to disk. See the examples section for example output generated by this callback.
Parameters: - printfreq : int >= 0
- Print the evaluation number every printfreq evaluations. Set to 0 to disable.
- names_to_log : Sequence[str]
- Names of the Data Sets (as set by the
Optimization
) to log.
Defaults to all Data Sets. - path : str
- Base name of the path where the data should be written to.
Defaults to theOptimization
workdir. - writefreq_history : int >= 0
- Write the history.dat every n calls.
Defaults to writing on every improvement of the loss function value. Set to zero to disable logging. - writefreq_datafiles : int >= 0
- Write the predictions.dat + combinations.dat every n calls.
Defaults to writing on every improvement of the loss function value. Set to zero to disable logging. - writefreq_bestparams : int >= 0
- Write the best_params every n calls.
Defaults to writing on every improvement of the loss function value. Set to zero to disable logging.
- *history.dat: The (fx,x) value pair at every evaluation (Note: Unscaled x,
3.10.2. Timeout¶
-
class
Timeout
(timeout_seconds, verbose=True)¶ Stop the optimization after timeout_seconds seconds. If verbose, prints a message when invoked.
3.10.3. Target Value¶
-
class
TargetValue
(min_fx, verbose=True)¶ Stop the optimization when the training set loss is less or equal to min_fx. If verbose, prints a message when invoked.
3.10.4. Maximum Iterations¶
-
class
MaxIter
(max_iter, verbose=True)¶ Stop the optimization after max_iter evaluations. If verbose, prints a message when invoked.
3.10.5. Early Stopping¶
-
class
EarlyStopping
(watch='trainingset', patience=0, verbose=True)¶ Stop the optimization if the data set defined in watch does not improve after patience iterations. If verbose, prints a message when invoked.
3.10.6. Stopfile¶
-
class
Stopfile
(fname='STOP', frequency=10, verbose=True)¶ Every frequency evaluations, check if a file named fname exists and stop the optimization if it does. Note that paths will be relative to the optimization directory.
3.10.7. Time per Evaluation¶
-
class
TimePerEval
(printfrequency=100, watch=None, workers=1, moving_average=50)¶ Print the average evaluation time of a new parameter set x every printfrequency iterations.
3.10.8. Load Average¶
-
class
LoadAvg
(fname, frequency=20)¶ Wrapper around psutil.getloadavg(), printing the otput to fname. Requires psutil version >= 5.6.2.
Note that when using relative file paths, the location will be relative to the optimization direvtory.
3.10.9. User-Defined Callbacks¶
The abstract Callback
class allows the user to define custom optimization hooks.
We will demonstrate the implementation of EarlyStopping
as an example below.
from scm.params import Callback
class EarlyStopping(Callback):
def __init__(self, patience=0):
self.patience = patience
self.count = 0
self.fxmin = float('inf')
def __call__(self,
fx : float,
x : Sequence[float],
name : str,
ncalled : int,
interface : Type[BaseParameters],
dataset : DataSet,
contrib : dict,
results : Union[AMSResults, AMSWorkerResults]
):
'''
Callbacks operate on **ALL** Data Sets that are evaluated at every optimization step,
meaning there could be more than one Data Set involved: This is for example the case when splitting
into a training and a validation set.
You can filter which Data Sets the callback operates on by checking the passed `name` argument --
those are always unique per Optimization instance.
'''
if name == 'validationset': # Only apply to the validation set
if np.isnan(fx): # nan means no evaluation for this call
return
if fx < self.fxmin:
self.count = 0 # Reset the counter if we improved
self.fxmin = fx # Adjust the best fx value
else:
self.count += 1 # Patience counter
ret = self.count > self.patience # Do we need to stop?
return ret
3.10.10. Callback API¶
-
class
Callback
¶ Abstract base class for callbacks
-
__call__
(evalret: scm.params.core.opt_components.EvaluatorReturn) → Any¶ This method will be called by the optimizer at the end of every step.
Parameters: - evalret : EvaluatorReturn (named tuple)
A named tuple returned by scm.params.core.opt_components.EvaluatorReturn. The tuple unpacks to
(fx, x, name, ncalled, interface, dataset, residuals, contrib, time)
. The names above also double as instance variables (e.g., fx can be accessed with evalret.fx).- fx : float
- Loss function value of x
- x : Sequence[float]
- The current set of parameters suggested by the optimizer. (real, not scaled)
- name : str
- Name of the Date Set as set by the
Optimization
class. Can be ‘trainingset’, ‘validationset’ and ‘datasetXX’ (wherte XX is an int) by default - ncalled : int
- The number of times this Data Set has been evaluated
- interface : BaseParameters subclass
- The interface that was used for this evaluation of the data set
- dataset : DataSet
- A tuple of
DataSet
and the last evaluation’s (non flattened) residuals vector - residuals : Lisd[1d-array]
- A list of 1d numpy arrays holding the residuals to each data set entry such that \(r=y-\hat{y}\). See Data Set for more information.
- contrib : List
- List of per entry contributions to the loss function value (see also
scm.params.core.dataset.DataSet.evaluate()
). - time : float
- Wall time (in seconds) this evaluation took
Returns: Any value other than None will be interpreted by the optimizer as a signal to stop the optimization process.
-
reset
()¶ This method should re-initialize the callback and will be called when a new
Optimization
instance is created containing this callback. It should reset the callback to it’s initial state, making the same instance available for multipleOptimization
instances (e.g.: In case ofTimeout
a reset of the same instance is necessary to reset the timer).
-
on_end
()¶ This method will be called once the optimization is complete
-