BaseNNConstr#

class gurobi_ml.modeling.neuralnet.neural_net.BaseNNConstr(gp_model, predictor, input_vars, output_vars, **kwargs)#

Bases: AbstractPredictorConstr

Base class for inserting a regressor based on a neural-network/tensor into Gurobi.

This only supports sequential neural networks.

The bracket operator can be used to iterate over the layers of the layers of the network. It will give access to the modeling object of the corresponding layer.

Attributes:

constrs: List of linear constraints added.
default_name: Default base name base used for automatic name generation.
genconstrs: List of general constraints added.
gp_model: Access gurobipy model this is a part of.
input: Input variables of embedded predictor.
input_values: Values for the input variables if a solution is known.
layers: Access models for successive layers of the network
output: Output variables of embedded predictor.
output_values: Values for the output variables if a solution is known.
qconstrs: List of quadratic constraints added.
sos: List of SOS constraints added.
vars: List of variables added.

Methods

`get_error`(eps)	Returns error in Gurobi's solution with respect to prediction from input.
`print_stats`([abbrev, file])	Print statistics about submodel created.
`remove`()	Remove from gp_model everything that was added to embed predictor.

abstract get_error(eps)#

Returns error in Gurobi’s solution with respect to prediction from input.

Returns:: error – Assuming that we have a solution for the input and output variables x, y. Returns the absolute value of the differences between predictor.predict(x) and y. Where predictor is the regression model represented by this object.
Return type:: ndarray of same shape as gurobi_ml.modeling.base_predictor_constr.AbstractPredictorConstr.output
Raises:: NoSolution – If the Gurobi model has no solution (either was not optimized or is infeasible).

print_stats(abbrev=False, file=None)#

Print statistics about submodel created.

Parameters:: file (None, optional) – Text stream to which output should be redirected. By default sys.stdout.

remove()#: Remove from gp_model everything that was added to embed predictor.

property constrs#: List of linear constraints added.

property default_name#: Default base name base used for automatic name generation.

property genconstrs#: List of general constraints added.

property gp_model#: Access gurobipy model this is a part of.

property input#

Input variables of embedded predictor.

Returns:: output
Return type:: gurobipy MVar.

property input_values#

Values for the input variables if a solution is known.

Returns:: output_value
Return type:: ndarray or pandas dataframe with values
Raises:: NoSolution – If the Gurobi model has no solution (either was not optimized or is infeasible).

property layers#: Access models for successive layers of the network

property output#

Output variables of embedded predictor.

Returns:: output
Return type:: gurobipy MVar.

property output_values#

Values for the output variables if a solution is known.

Returns:: output_value
Return type:: ndarray or pandas dataframe with values
Raises:: NoSolution – If the Gurobi model has no solution (either was not optimized or is infeasible).

property qconstrs#: List of quadratic constraints added.

property sos#: List of SOS constraints added.

property vars#: List of variables added.