w,v从哪里来

[qqxze]

您好， # FM part 1st order layer
self.w = self.params.get('w', grad_req='write', shape=(self._feature_dim, 1))

        # FM part 2nd order vector
        self.v = self.params.get('v', grad_req='write', shape=(self._feature_dim, k_units))

def hybrid_forward(self, F, x,w, v):
    x_broadcast = x.reshape((-1, self._feature_dim, 1))
    # FM first order
    f1 = F.broadcast_mul(w, x_broadcast)
    f1 = F.sum(f1, axis=2)

    # FM 2nd order
    xv = F.broadcast_mul(v, x_broadcast)
    f2 = 0.5 * (F.square(F.sum(xv, axis=1)) - F.sum(F.square(xv), axis=1))
    f2 = F.Dropout(f2,p_drop)
    # DNN
    xv = F.flatten(xv) #shape=[-1,self._feature_dim*self._k_units])
    #print (xv.shape)
    for i in range(len(self._dnn_layers)):
        xv = self.denselayers[i](xv)
        xv = self.droplayers[i](xv)
    # concate all parts
    y = F.concat(xv, f1, f2, dim=1)
    y = self.densefinal(y)
    return y

在前馈网络中，w,v是由网络参数入口获得，但是在您train里边，net()里边送入的参数只有 x，没有w,v 并且self.w,self.v 您也并没有用到。所以，请问前馈网络中的w v到底是从哪里来的呐？ 谢谢啦

[xiehust]

v,w就是这2个变量，这个代码有点老了，当时mxnet可能还不太完善，直接用self.w,self.v貌似不行，反而写出hybrid_forward(self, F, x,w, v)能跑通。
self.w = self.params.get('w', grad_req='write', shape=(self._feature_dim, 1))
self.v = self.params.get('v', grad_req='write', shape=(self._feature_dim, k_units))

[xiehust]

official tutorials http://mxnet.incubator.apache.org/api/python/docs/tutorials/packages/gluon/blocks/custom_layer_beginners.html

`class NormalizationHybridLayer(gluon.HybridBlock):
def init(self, hidden_units, scales):
super(NormalizationHybridLayer, self).init()

    with self.name_scope():
        self.weights = self.params.get('weights',
                                       shape=(hidden_units, 0),
                                       allow_deferred_init=True)

        self.scales = self.params.get('scales',
                                  shape=scales.shape,
                                  init=mx.init.Constant(scales.asnumpy()),
                                  differentiable=False)

def hybrid_forward(self, F, x, weights, scales):
    normalized_data = F.broadcast_div(F.broadcast_sub(x, F.min(x)), (F.broadcast_sub(F.max(x), F.min(x))))
    weighted_data = F.FullyConnected(normalized_data, weights, num_hidden=self.weights.shape[0], no_bias=True)
    scaled_data = F.broadcast_mul(scales, weighted_data)
    return scaled_data`