layg.learner.Learner

class layg.learner.Learner(true_func: Callable[[numpy.ndarray], numpy.ndarray], emulator_class)

A class that contains logic for learning as you go

This class does not contain any emulation but should be constructed with an emulator containing emulation logic. The emulator must be a subclass of BaseEmulator, implementing two methods, set_emul_func and set_emul_error_func, that set the respective functions.

Attributes:

emulator_class : BaseEmulator

The type of emulator used.

emulator : BaseEmulator

An instance of the class emulator_class. This is where the heavy lifting goes on.

true_func : Callable

The function which is emulated

frac_err_local : float

Maximum fractional error in emulated function. Calls to emulation function that exceed this error level are evaluated exactly instead. Default: 1.0

abs_err_local : float

Maximum absolute error allowed in emulated function. Calls to emulation function that exceed frac_err_local but are lower than abs_err_local are emulated, rather than exactly evaluated. FIXME: this doesn’t happen Default: 0.05

output_err : bool

Whether to output an error estimate. Set to False if you do not want the error to be an output of the emulated function. Set to True if you do. Default: False

trained : bool

Whether the emulator has been trained

used_train_x : List[np.ndarray]

used_train_y : List[np.ndarray]

Values from the true function that were used last time the emulator was trained

batch_train_x : List[np.ndarray]

batch_train_y : List[np.ndarray]

Values from the true function that have not yet been used to train the emulator

init_train_thresh : int

Number of points to accumulate before training the emulator

frac_cv : float

Fraction of training set to use for error modelling The default value of 0.5 means that the prediction and the error are estimated off the same amount of data.

Methods

__call__(self, x)	The method that is executed when the wrapped function is called
emulation_is_valid(self, val, err)	Check if an emulated value is valid and likely accurate
eval_true_func(self, x)	Wrapper for evaluating true function
split_CV(self, xdata, ydata, frac_cv)	Splits a dataset into a cross-validation and training set.
train(self, x_train, y_train)	Train a ML algorithm to replace true_func: X –> Y.

__init__(self, true_func: Callable[[numpy.ndarray], numpy.ndarray], emulator_class)

Constructor for Learner class

Parameters:

true_func : Callable

Function to be emulated

emulator_class : BaseEmulator

The emulator class to be used

Methods

__init__(self, true_func, numpy.ndarray], …)	Constructor for Learner class
emulation_is_valid(self, val, err)	Check if an emulated value is valid and likely accurate
eval_true_func(self, x)	Wrapper for evaluating true function
split_CV(self, xdata, ydata, frac_cv)	Splits a dataset into a cross-validation and training set.
train(self, x_train, y_train)	Train a ML algorithm to replace true_func: X –> Y.