sciope.models package

Submodules

sciope.models.ann_regressor module

Artificial Neural Network (ANN) Regression Surrogate Model

class sciope.models.ann_regressor.ANNModel(use_logger=False)

Bases: sciope.models.model_base.ModelBase

We use the sklearn MLP Regressor implementation here.

predict(xt)

Predict unseen data using the trained model

xtnd-array

unseen data to be predicted

vector

predictions

train(inputs, targets)

Train the ANN model given the data

inputsnd-array

independent variables

targetsvector

dependent variable

sciope.models.gp_regressor module

Gaussian Process Regression Surrogate Model using sklearn

class sciope.models.gp_regressor.GPRModel(use_logger=False)

Bases: sciope.models.model_base.ModelBase

We use the sklearn GP Regressor implementation here.

predict(xt)

Predict unseen data using the trained model. GP returns the mean and variance of prediction, so handle it accordingly while using predict.

xtnd-array

unseen data to be predicted

tuple of vectors

predictions, prediction variance

train(inputs, targets)

Train the GP model given the data

inputsnd-array

independent variables

targetsvector

dependent variable

sciope.models.label_propagation module

Semi-supervised Label Propagation Surrogate Model

class sciope.models.label_propagation.Bounds(xmax, xmin)

Bases: object

class sciope.models.label_propagation.LPModel(kernel='rbf', alpha=0.7, gamma=0.1, learning_rate=1.0, use_logger=False)

Bases: sciope.models.model_base.ModelBase

We use the sklearn Label Spreading implementation here.

objective(x)

Objective function for hyper-parameter selection/evaluation

x : hyper-parameter under test - gamma

float

a measure directly proportional to entropy

optimize(min_, max_, niter=10, stepsize=0.1)

hyper-parameter optimization using basinhopping

min_ : lower bound for parameter search max_ : upper bound for parameter search niter : number of allowed optimization iterations stepsize : displacement step size

vector

optimized hyper-parameters

predict(xp)

Predict unseen data using the trained model

xpnd-array

unseen data to be predicted

vector

predictions

train(inputs, targets, min_=0.01, max_=30, niter=10, stepsize=0.1)

Train the LP model given the data

inputsnd-array

independent variables

targetsvector

dependent variable

minfloat

[]

maxfloat

[]

niterint

number of training iterations

stepsizefloat

[]

class sciope.models.label_propagation.RandomDisplacementBounds(xmax, xmin, stepsize=0.5)

Bases: object

random displacement with bounds

sciope.models.label_propagation.get_average_label_entropy(label_distribution)

sciope.models.label_propagation.get_label_entropies(label_distribution)

sciope.models.model_base module

Surrogate Model Base Class

class sciope.models.model_base.ModelBase(name, use_logger=False)

Bases: object

Base class for surrogate models. Must not be used directly! Each model type must implement the methods described herein:

• ModelBase.train(x,y)

• ModelBase.predict(xt)

The following variables are available to derived classes:

• self.x (training inputs)

• self.y (training targets)

• self.my (mean for scaling)

• self.sy (std dev for scaling)

• self.n (num training points)

abstract predict(test_input)

Sub-classable method for predicting test data. Each derived class must implement.

scale_training_data(x, y)

pre-process training data

xnd-array

inputs or independent variables

ynd-array

output or dependent variable

abstract train(inputs, targets)

Sub-classable method for training a given surrogate. Each derived class must implement.

sciope.models.svm_regressor module

Support Vector Machine Regression Surrogate Model

class sciope.models.svm_regressor.SVRModel(use_logger=False)

Bases: sciope.models.model_base.ModelBase

We use the sklearn SVM implementation here.

predict(xt)

Predict unseen data using the trained model

xtnd-array

unseen data to be predicted

vector

predictions

train(inputs, targets)

Train the SVR model given the data

inputsnd-array

independent variables

targetsvector

dependent variable

tune_parameters(x, y, nfolds)

Tune hyper-parameters of the model

x : inputs or independent variables y : output or dependent variable nfolds : number of cross-validation folds

vector

pseudo-optimal parameters

Module contents