twinlab.Emulator.learn#

Emulator.learn(dataset, inputs, outputs, num_loops, num_points_per_loop, acq_func, simulation, train_params=<twinlab.params.TrainParams object>, recommend_params=<twinlab.params.RecommendParams object>, verbose=False)[source]#

Perform active learning to improve an emulator on the twinLab cloud.

Active learning is a method that can identify and utilise the most informative data points to add to an emulator in order to reduce the number of measurements to be taken or simulations that are required. Using active learning can result in a more accurate model, trained with less data. The primary difference between this method and Emulator.recommend is that in this method the emulator is trained on the new data points that are suggested in an active loop. This way, new data can be used to update an emulator until the desired level of accuracy is achieved. This can be done using either the "optmise" or "explore" acquisition functions. The emulator can therefore be updated with the objective of either finding the point of maximum output or reducing the overall uncertainty in the emulator.

Parameters:

dataset (Dataset) – twinLab dataset object which contains the initial training data for the emulator.
inputs (list[str]) – List of input column names in the training dataset.
outputs (list[str]) – List of output column names in the training dataset.
num_loops (int) – Number of loops to run the learning process.
num_points_per_loop (int) – Number of points to sample in each loop.
acq_func (str) – Specifies the acquisition function to be used when recommending new points: either "explore" or "optimise".
simulation (Callable) – A function that takes in a set of inputs and generates the outputs (for example, a simulator for the data generating process).
train_params (TrainParams, optional) – A parameter configuration that contains optional training parameters.
recommend_params (RecommendParams, optional) – A parameter configuration that contains optional recommendation parameters.
verbose (bool, optional) – Display detailed information about the operation while running.

Return type:

None

Examples

emulator = tl.Emulator(id=emulator_id)
dataset = tl.Dataset(dataset_id)
emulator.learn(
    dataset=dataset,
    inputs=["X"],
    outputs=["y"],
    num_loops=3,
    num_points_per_loop=5,
    acq_func="explore",
    simulation=my_simulator,
)