MLPRegressor

#include <Skigen/NeuralNetwork>

template <typename Scalar = double>
class Skigen::MLPRegressor(hidden_layer_sizes={100}, activation=MLPActivation::ReLU, solver=MLPSolver::Adam, alpha=1e-4, learning_rate_init=1e-3, max_iter=200, tol=1e-4, batch_size=0, random_state=std::nullopt, beta_1=0.9, beta_2=0.999, epsilon=1e-8)

Multi-layer perceptron regressor with squared-error loss.

Mirrors sklearn.neural_network.MLPRegressor.

Architecture: an arbitrary stack of fully-connected hidden layers with the chosen activation, followed by a linear output layer. Trained by mini-batch SGD or Adam on the squared-error loss with L2 weight regularisation (alpha).

---

---

Attributes:

• hidden_layer_sizes : const std::vector< int > &

• activation : MLPActivation

• solver : MLPSolver

• n_iter_run : int

• loss : Scalar

• coefs : const std::vector< MatrixType > &

• intercepts : const std::vector< VectorType > &

• n_outputs : int

---

Methods

SKIGEN_PARAMS()

---

fit_multi(X, Y)

Fit on multiple target columns.

---

predict(X)

---

predict_multi(X)

Predict multiple targets (returns n_samples × n_targets matrix).

---

score(X, y)

---