Tensors

class ConstantTensor(*args: Any, **kwargs: Any)

Bases: Module

Pytorch module to implement a constant function (always one).

class GenericBasisTensor(*args: Any, **kwargs: Any)

Bases: Module

Abstract Pytorch module to implement a generic basis function. All the basis function generated by this module are

__init__(mu, scale, dim=None, normalized=False)

Constructor.

Parameters:

• mu (np.ndarray) – centers of the gaussian RBFs;

• scale (np.ndarray) – scales for the RBFs;

• dim (np.ndarray, None) – list of dimension to be considered for the computation of the features. If None, all dimension are used to compute the features;

• normalized (bool, False) – whether the features need to be normalized to sum to one or not;

static _convert_to_scale(w)

Converts width of a basis function to scale

Parameters:

w (np.ndarray) – array of widths of basis function for every dimension

Returns:

The array of scales for each basis function in any given dimension

classmethod is_cyclic()

Method used to change the basis generation in case of cyclic features. :returns: Whether the space we consider is cyclic or not.

classmethod generate(n_centers, low, high, dimensions=None, eta=0.25, normalized=False)

Factory method that generates the list of dictionaries to build the tensors representing a set of uniformly spaced radial basis functions with eta overlap.

Parameters:

• n_centers (list) – list of the number of radial basis functions to be used for each dimension;

• low (np.ndarray) – lowest value for each dimension;

• high (np.ndarray) – highest value for each dimension;

• dimensions (list, None) – list of the dimensions of the input to be considered by the feature. The number of dimensions must match the number of elements in high and low;

• eta (float, 0.25) – percentage of overlap between the features;

• normalized (bool, False) – whether the features need to be normalized to sum to one or not.

Returns:

The tensor list.

class GaussianRBFTensor(*args: Any, **kwargs: Any)

Bases: GenericBasisTensor

static _convert_to_scale(w)

Converts width of a basis function to scale

Parameters:

w (np.ndarray) – array of widths of basis function for every dimension

Returns:

The array of scales for each basis function in any given dimension

class VonMisesBFTensor(*args: Any, **kwargs: Any)

Bases: GenericBasisTensor

classmethod is_cyclic()

Method used to change the basis generation in case of cyclic features. :returns: Whether the space we consider is cyclic or not.

static _convert_to_scale(w)

Converts width of a basis function to scale

Parameters:

w (np.ndarray) – array of widths of basis function for every dimension

Returns:

The array of scales for each basis function in any given dimension

class RandomFourierBasis(*args: Any, **kwargs: Any)

Bases: Module

Class implementing Random Fourier basis functions. The value of the feature is computed using the formula:

\[\sin{\dfrac{PX}{\nu}+\varphi}\]

where \(X\) is the input, \(P\) is a random weights matrix, \(\nu\) is the bandwidth parameter and \(\varphi\) is a bias vector.

These features have been presented in:

“Towards generalization and simplicity in continuous control”. Rajeswaran A. et Al.. 2017.

__init__(P, phi, nu)

Constructor.

Parameters:

• P (np.ndarray) – weights matrix, every weight should be drawn from a normal distribution;

• phi (np.ndarray) – bias vector, every weight should be drawn from a uniform distribution in the interval :math: [-pi, pi);

• nu (float) – bandwidth parameter, it should be chosen approximately as the average pairwise distances between different observation vectors.

static generate(nu, n_output, input_size, use_bias=True)

Factory method to build random fourier basis. Includes a constant tensor into the output.

Parameters:

• nu (float) – bandwidth parameter, it should be chosen approximately as the average pairwise distances between different observation vectors.

• n_output (int) – number of basis to use;

• input_size (int) – size of the input.

Returns:

The list of the generated fourier basis functions.