Method fit

airline/ann.py:41–116

(self, X, Y, activation=T.tanh, learning_rate=1e-3, mu=0.5, reg=0, epochs=5000, batch_sz=None, print_period=100, show_fig=True)

        self.hidden_layer_sizes = hidden_layer_sizes

    def fit(self, X, Y, activation=T.tanh, learning_rate=1e-3, mu=0.5, reg=0, epochs=5000, batch_sz=None, print_period=100, show_fig=True):
        X = X.astype(np.float32)
        Y = Y.astype(np.float32)

        # initialize hidden layers
        N, D = X.shape
        self.hidden_layers = []
        M1 = D
        count = 0
        for M2 in self.hidden_layer_sizes:
            h = HiddenLayer(M1, M2, activation, count)
            self.hidden_layers.append(h)
            M1 = M2
            count += 1
        W = np.random.randn(M1) / np.sqrt(M1)
        b = 0.0
        self.W = theano.shared(W, 'W_last')
        self.b = theano.shared(b, 'b_last')

        if batch_sz is None:
            batch_sz = N

        # collect params for later use
        self.params = [self.W, self.b]
        for h in self.hidden_layers:
            self.params += h.params

        # for momentum
        dparams = [theano.shared(np.zeros(p.get_value().shape)) for p in self.params]

        # set up theano functions and variables
        thX = T.matrix('X')
        thY = T.vector('Y')
        Yhat = self.forward(thX)

        rcost = reg*T.mean([(p*p).sum() for p in self.params])
        cost = T.mean((thY - Yhat).dot(thY - Yhat)) + rcost
        prediction = self.forward(thX)
        grads = T.grad(cost, self.params)

        # momentum only
        updates = [
            (p, p + mu*dp - learning_rate*g) for p, dp, g in zip(self.params, dparams, grads)
        ] + [
            (dp, mu*dp - learning_rate*g) for dp, g in zip(dparams, grads)
        ]

        train_op = theano.function(
            inputs=[thX, thY],
            outputs=[cost, prediction],
            updates=updates,
        )

        self.predict_op = theano.function(
            inputs=[thX],
            outputs=prediction,
        )