Temat pracy

Zastosowanie sztucznych sieci neuronowych
do diagnostyki stanów ostrego brzucha

Application of artificial neural networks
to the diagnosis of surgical abdomen states

Prowadzący pracę

prof. dr hab. inż.

Marek Kurzyński

Technologie

Aplikacja

Poboczne

API

Schemat warstwy

class Layer:
    def __init__(self, shape, activation='sigmoid'):
        ...

    def feedforward(self, x: np.ndarray) -> np.ndarray:
        ...

    def calc_delta(self, d: np.ndarray = None):
        ...

    def calc_gradient(self):
        ...

    def update_weights(self):
        ...

Architektura sieci

def input_data(shape: Tuple[Optional[int], int]) -> Layer:
    ...


def fully_connected(incoming: Layer, n_units: int,
                    activation='relu') -> Layer:
    ...


net = input_data(shape=(None, x_train.shape[1]))
net = fully_connected(net, 24, activation='sigmoid')
net = fully_connected(net, 16, activation='sigmoid')
net = fully_connected(net, 12, activation='sigmoid')
net = fully_connected(net, 8, activation='sigmoid')

Schemat modelu

class Model:
    def __init__(self, network: Layer):
        ...

    def fit(self, X_inputs: np.ndarray, Y_targets: np.ndarray,
            validation_set: Tuple[np.ndarray, np.ndarray] = None,
            learning_rate=None, n_epoch=10, batch_size=64,
            shuffle=False, train_file='train.json'):
        ...

    def predict(self, x: np.ndarray) -> np.ndarray:
        ...

    def load(self, model_file: str):
        ...

    def save(self, model_file: str):
        ...

Uczenie modelu

model_file = 'model.json'
model = Model(net)
model.fit(x_train, y_train,
          validation_set=(x_test, y_test),
          n_epoch=30,
          batch_size=10)
model.save(model_file)
model.load(model_file)

Plan prac