BFS
====

.. _BFS ::

::
    
    BFS(n_attr, pref_vector1, pref_vector2. n_hidden=[100,200], n_output)
    
BFS is a subclass of the keras Model module. However, BFS is based on biobjective gradient descent.

The implementation of the BFS module is based on the keras Model module. However, 
BFS has two objective functions in the network: the classification error (cross entropy)
and the sparse group lasso. 

Returns a sparse model with features ranked based on the importance of each feature. 

Arguments
----------

n_attr: int
    Number of attributes in the data

pref_vector1: array
    Array containing the coordinates of the first preference vector delimiting the concerned region

pref_vector2: array
    Array containing the coordinates of the second preference vector delimiting the concerned region
    
n_hidden: array, default = [100,200]
    Array containing the number of neurons per hidden layer. The size of the array indicates the number of hidden layers.
    Default: 2 hidden layers of 100 and 200 neurons respectively. 

n_output: int, default = 2
    Number of classes in the data. By default, the number of classes is 2. 
    
    
Methods
--------

**Compile**

::

    compile(init_optimizer='Adam', train_optimizer='Adam', init_learning_rate=0.01,
            train_learning_rate=0.01, metrics = ['accuracy'], **kwargs)

Configures the model for training.

**Parameters:**

init_optimizer: String, default: 'Adam'
    Name of optimizer to be used for the initialization step. Default optimizer is Adam. 
    
train_optimizer: String, default: 'Adam'
    Name of optimizer to be used for the training step. Default optimizer is Adam.
    
init_learning_rate: float, default: 0.01
    Learning rate to be used for optimization in the initialization step. Default is 0.01.

train_learning_rate: float, default: 0.01
    Learning rate to be used for optimization in the training step. Default is 0.01.
    
metrics: list, default: ['accuracy']
    List of metrics to be evaluated by the model during training and testing. 
    See ``keras.metrics`` for more details on the accepted metrics. Default is the accuracy metric.

--------

**fit**

::

    fit(X, y, validation_data, callbacks, batch_size = 32, epochs = 500, **kwargs)

Trains the model for a fixed number of epochs (1 epoch = whole dataset iteration). 

**Parameters**

X: array or tensor 
    Input data. It can be either a NumPy array, or a list of arrays (in case of multiple inputs), or a tensor,
    or a list of tensors (again, in case of multiple inputs). 
    
y: array or tensor
    Target data. Just like `X`, it can be either array(s) or tensor(s). 
    
validation_split: float, default: 0.2
    Fraction of the data that will be used as training. 

validation_data: tuple
    Data on which to evaluate the loss and any model metrics at the end of each epoch. It can be a tuple of NumPy arrays 
    or tensors. `validation_data` if designated will override `validation_split`. 
    
callbacks: list
    List of callbacks to apply during training. In addition to the keras.callbacks.History callback, 
    BFS has the checkTrain callback that is also called automatically for switiching between the initialization and 
    training steps of the learning step. 

batch_size: int, default: 32
    Number of samples per gradient update. By default, the batch size is 32.
    
epochs: int, default: 500
    Number of epochs to train the model. One epoch is an iteration over the entire `x` and corresponding `y` data provided. 
    By default, the model will train 500 epochs. 

**Returns**

A `History` object. Its `History.history` attribute is a dictionary containing all training loss values, for both
losses, and metrics for all epochs as well all validation losses values and validation metrics when found.

--------

**predict**

::

    predict(x)
    
x: NumPy array or tensor
    Input data. It can be either a NumPy array, or a list of arrays (in case of multiple inputs), or a tensor,
    or a list of tensors (again, in case of multiple inputs). 
    
**Returns**

NumPy array(s) of predictions.

--------

**evaluate**

::

    evaluate(X, y, return_dict= False, **kwargs)
    
**Parameters**

X: array or tensor 
    Input data. It can be either a NumPy array, or a list of arrays (in case of multiple inputs), or a tensor,
    or a list of tensors (again, in case of multiple inputs). 
    
y: array or tensor
    Target data. Just like `X`, it can be either array(s) or tensor(s). 
    
return_dict: boolean, default: False
    If True, `evaluate` output returned as dictionary instead of list. 
    
**Returns**

A list of scalars containing both loss values and metrics if applicable. 

--------