isanet.neural_network¶

class isanet.neural_network.MLPClassifier(input_dim, out_dim, n_layer_units=[100], activation='relu', kernel_regularizer=0.0001, batch_size=None, max_epoch=1000, learning_rate=0.1, momentum=0.9, nesterov=False, sigma=None, early_stop=False, random_state=None, verbose=0)¶

Bases: isanet.neural_network.__BaseMLP

Multi-layer Perceptron classifier.

This model optimizes the MSE function using the stochastic gradient descent.

Parameters

input_dim (int, no default value mandatory parameter) – allows you to specify the number of inputs on the input layer
out_dim (int, no default value mandatory parameter) – allows you to specify the number of outputs on the output layer
hidden_layer_sizes (list, default=[100]) –
The ith element of the list represents the number of neurons in the ith hidden layer:
```
E.g. [20, 40, 60] means 3 hidden layers with 20, 40 and 60 neurons respectively.
```
activation ({'identity', 'sigmoid', 'tanh', 'relu'}, default='relu') –
Activation function available for the hidden layer.
- ’identity’, no-op activation, useful to implement linear bottleneck, returns f(x) = x
- ’sigmoid’, the logistic sigmoid function, returns f(x) = 1 / (1 + exp(-x)).
- ’tanh’, the hyperbolic tan function, returns f(x) = tanh(x).
- ’relu’, the rectified linear unit function, returns f(x) = max(0, x)
kernel_regularizer (float, default=0.0001) – Tikhonov regularization term, L2 penalty parameter.
batch_size (int, default='None') – Size of minibatches for the SGD optimizer. When set to “none”, the SGD will performe a full batch.
learning_rate (float, default=0.1) – The constant value that will be used by the SGD optimizer as learning rate
momentum (float, default=0.9) – Momentum for gradient descent update. Should be between 0 and 1.
nesterovs_momentum (boolean, default=True) – Whether to use Nesterov’s momentum. If the momentum == 0 this parameter is useless.
sigma (float, default=None) – Parameter of the Super Accelerated Nesterov’s momentum. If ‘nesterov’ is True and ‘sigma’ equals to ‘momentum’, then we have the simple Nesterov momentum. Instead, if ‘sigma’ is different from ‘momentum’, we have the super accelerated Nesterov.
max_epoch (int, default=1000) – It will set the Maximum number of Epoch for the SGD optimizer. The solver iterates until convergence (determined by ‘tol’) or this number of iterations.
early_stopping (bool or isanet.optimizer.EarlyStopping, default=False) – When set to False it will only use the max_epoch to finish training. Otherwise, an EarlyStopping type object has been passed and will stop training if the model goes overfitting after a number of consecutive iterations. See docs in optimizier module.
random_state (int, RandomState instance or None, default=None) – If int, random_state is the seed used by numpy random number generator; If None, the random number generator is the RandomState instance used by np.random.
verbose (bool, default=False) – Whether to print progress messages to stdout.

Notes

MLPClassifier provides a high-level interface capable of biting a neural network using the parameters passed to the class as hyper parameters.

It can also have a regularization term added to prevent overfitting.

Numpy arrays of floating point values are used to store all the data under the hood. this This matrix implementation allowed us to speed up the computation compared to an object-oriented structure, this was possible thanks to Numpy that is able to perform matrix operation in an efficient way by parallelizes each operation. Numpy use optimized math routines, written in C or Fortran, for linear algebra operation as: Blas, OpenBlas or Intel Math Kernel Library (MKL).

fit(self, X_train, Y_train, X_val, Y_val)¶: Fit the model to data matrix X_train and target(s) Y_train and evaluates it on the validation set (X_val, Y_val).

predict(self, X)¶: Predict using the multi-layer perceptron classifier.

get_params(self)¶: Returns the parameters of the multi-layer perceptron.

get_history(self)¶: Returns the history of the multi-layer perceptron.

get_weights(self)¶: Returns the weights of the multi-layer perceptron.

class isanet.neural_network.MLPRegressor(input_dim, out_dim, n_layer_units=[100], activation='relu', kernel_regularizer=0.0001, batch_size=None, max_epoch=1000, learning_rate=0.1, momentum=0.9, nesterov=False, sigma=None, early_stop=False, random_state=None, verbose=0)¶