modeling package

modeling.graphing module

The module graphing.py contains classes for building graphical structures of learned skills for Keyframe-based LfD.

class modeling.graphing.KeyframeGraph

NetworkX MultiDiGraph extended graph class for containing keyframes and their corresponding models.

cull_node(node)

Takes received node and removes it from the task graph

Parameters:

node: hashable networkx node name (usually int)

the node to be removed from network x graph

fit_models(observation_vectorizor)

Fits all models in graph. Expects that each keyframe node in graph has populated “observation” key as a list of Observation objects.

Parameters:

observation_vectorizor : function

Function that takes observations stored in graph and converts their data into vector form for use by model fitting function.

get_keyframe_sequence()

Provides the keyframe sequence in order.

Returns:

node_chain : list

List of keyframe/node sequence.

class modeling.graphing.ObservationClusterer

Clusters together observations by keyframe ID and gathers pertinent information regarding each keyframe. This will be used by a KeyframeGraph object.

assign_applied_constraints(cluster)

Assigns the applied constraints for the given keyframe cluster.

Parameters:

cluster : dict

Dictionary to assign the applied constraints..

assign_keyframe_type(cluster)

Assigns the keyframe type for the given keyframe cluster.

Parameters:

cluster : dict

Dictionary to assign the keyframe type.

cluster_observations_by_id(observations)

Takes in a the entirety of observations from all Demosntrations and groups them together by keyframe ID.

Parameters:

observations : list

List of the Observation objects to cluster.

Returns:

clusters : dict

Dictionary of clusters with top level keys the keyframe IDs and values another dictionary populated with ‘observations’ key with list of Observations.

generate_clusters(demonstrations)

Generates the clustered Observations from a list of Demonstrations.

Parameters:

demonstrations : list

List of the Demonstration objects from which labeled Observations.

Returns:

clusters : dict

Dictionary of clusters with top level keys the keyframe IDs and values another dictionary populated with pertinent information associated with each keyframe (i.e. keyframe type, applied constraints etc,.)

modeling.models module

Wrappers around various models

class modeling.models.GaussianMixtureModel(n_components=5, means_init=None)

Wrapper class for Scikit Learn’s Gaussian Mixture Model.

Attributes:

n_components : int

Number of clusters for KMeans

model : GaussianMixture

Wrapped class model.

fit(train_X)

Wrapper method for fit() method of GMM model.

Parameters:

train_X : {array-like, sparse matrix}, shape = [n_samples, n_features]

generate_samples(n_samples)

Generates the random samples according to the fitted distrubution.

Returns:

list

List of numpy arrays of randomly generated observations.

score_samples(X)

Predicts the log liklihood score of the samples in X.

Parameters:

X : {array-like, sparse matrix}, shape = [n_samples, n_features]

class modeling.models.KDEModel(kernel='gaussian', bandwidth=0.001)

Wrapper class for Scikit Learn’s Kernel Density Estsimation model.

Attributes:

model : KernelDensity

Wrapped class model.

fit(train_X)

Wrapper method for fit() method of Kernel Density model.

Parameters:

train_X : {array-like, sparse matrix}, shape = [n_samples, n_features]

generate_samples(n_samples)

Generates the random samples according to the fitted distrubution.

Returns:

list

List of numpy arrays of randomly generated observations.

score_samples(X)

Predicts the log liklihood score of the samples in X.

Parameters:

X : {array-like, sparse matrix}, shape = [n_samples, n_features]

class modeling.models.KMeansModel(n_clusters=5)

Wrapper class for Scikit Learn’s KMeans clustering.

Attributes:

n_clusters : int

Number of clusters for KMeans

model : KMeans

Wrapped class model.

fit()

Wrapper method for fit() method of kmeans model.

get_clusters(train_X)

Generates the raw samples associated with each cluster.

Parameters:

train_X : {array-like, sparse matrix}, shape = [n_samples, n_features]

Returns:

cluster_data : dict

Dictionary of clusters.

score_samples(X)

Predicts which cluster each of the samples in X belongs.

Parameters:

X : {array-like, sparse matrix}, shape = [n_samples, n_features]

modeling.sampling module

The module sampling.py contains classes for sampling and ranking points from Keyframe models.

class modeling.sampling.KeyframeSampler(analyzer, data_converter)

Sampling class that uses model representing a keyframe to sample points.

Attributes:

analyzer : object

Analysis object that evaluates sampled points for their validity (constraint satisfcation).

data_converter : func

Function to convert a raw sample into an Observation object for use by the analyzer.

generate_n_valid_samples(model, constraint_ids, n=100)

Parameters:

model : object

Model object used to generate raw samples.

constraint_ids : list

List of constraint IDs required for validity.

n : int

Number of valid samples to generate.

Returns:

raw_samples : list

List of valid raw samples generated from the model.

generate_raw_samples(model, num_of_samples)

Parameters:

model : object

Model object used to generate raw samples.

num_of_samples : int

Number of samples to generate.

Returns:

raw_samples : list

List of raw samples generated from the model.

rank_samples(model, samples)

Parameters:

model : object

Model object used to generate raw samples.

samples : list

List of samples to rank according to their score as measured by the model.

Returns:

rank_sorted_sampled : list

List of rank (according to model scoring function) sorted samples (descending order).

modeling.visualization module

The visualization.py module constains classes for viewing sample poitns and models used in Cairo LfD.

class modeling.visualization.GaussianMixtureModelViewer(gmm, observation_vectors)

Class for viewing observation vector representaions within Gaussian Mixture Model.

Attributes:

kmm : KMeansModel

Model to visualize.

observation_vectors : list

List of vectorized Observations.

view_2D_gaussians(x_index, y_index)

Generates 2D graph of distribution components of the of Gaussian Mixture model.

Parameters:

x_index : int

Index of sample point to represent the x-axis value.

y_index : int

Index of sample point to represent the y-axis value.

class modeling.visualization.KMeansModelViewer(kmm, observation_vectors)

Class for viewing observation vector representation with KMeansModel.

Attributes:

kmm : KMeansModel

Model to visualize.

observation_vectors : list

List of vectorized Observations.

view_3D_clusters(x_index, y_index, z_index)

Generates 3D graph of the clusters of KMeans model

Parameters:

x_index : int

Index of sample point to represent the x-axis value.

y_index : int

Index of sample point to represent the y-axis value.

z_index : int

Index of sample point to represent the z-axis value.

class modeling.visualization.SamplePointViewer

Class for viewing samples points via Matplotlib’s pyplot.

view_3D_scatter(sample_points, x_index, y_index, z_index)

Generates 3D graph according of the passed in sample points.

Parameters:

sample_points : list

List of numpy arrays (usually observations)

x_index : int

Index of sample point to represent the x-axis value.

y_index : int

Index of sample point to represent the y-axis value.

z_index : int

Index of sample point to represent the z-axis value.