Module propinfer.model

Classes

class LinReg (label_col, hyperparams=None)

A linear regression based model

Args

label_col

the index of the column to be used as Label

hyperparams : dict of DictConfig

hyperperameters for the Model Accepted keywords: max_iter (default = 100), normalise (default=False)

Ancestors

• Model

Subclasses

• LogReg

Inherited members

• Model:
  • fit
  • parameters
  • predict
  • predict_proba

class LogReg (label_col, hyperparams)

A logistic regression based model

Args

label_col

the index of the column to be used as Label

hyperparams : dict of DictConfig

hyperperameters for the Model Accepted keywords: max_iter (default = 100), normalise (default=False)

Ancestors

• LinReg
• Model

Inherited members

• LinReg:
  • fit
  • parameters
  • predict
  • predict_proba

class MLP (label_col, hyperparams)

A Multi-Layer Perceptron based model, for either regression or classification

Args

label_col

the index of the column to be used as Label

hyperparams : dict of DictConfig

hyperperameters for the Model Accepted keywords: input_size (mandatory), n_classes (mandatory, performs regression if is 1), layers (default=[64,16]), epochs (default=20), learning_rate (default=1e-1), weight_decay (default=1e-2), batch_size (default=32), normalise (default=False)

Ancestors

• Model

Inherited members

• Model:
  • fit
  • parameters
  • predict
  • predict_proba

class Model (label_col, normalise)

An abstract class to be extended to represent the models that will be attacked.

Args

label_col

the index of the column to be used as Label

normalise : bool

whether to normalise data before fit/predict

Subclasses

• LinReg
• MLP

Methods

def fit(self, data)

Fits the model according to the given data

Args

data

DataFrame containing all useful data

Returns: Model, the model itself

def parameters(self)

Returns the model's parameters.

• If the model has only one layer, or is not a DNN, as a numpy array.
• If the model has multiple layers without biases, as a list of numpy arrays representing each layer.
• If the model has multiple layers with weights and biases, arrays of the corresponding weights and biases are grouped in a list, with weights going before biases.

Returns: the model's parameters

def predict(self, data)

Makes predictions on the given data

Args

data

DataFrame containing all useful data

Returns: np.array containing predictions

def predict_proba(self, data)

Outputs prediction probability scores for the given data

Args

data

DataFrame containing all useful data

Returns:np.array containing probability scores