def _build_default_interactions(self, train=True):
self._add_module(
'interaction',
PointwiseGeCE(
user=self._get_module('user_vec', train=train).get_outputs()[0],
item=self._get_module('item_vec', train=train).get_outputs()[0],
item_bias=self._get_module('item_bias', train=train).get_outputs()[0],
l2_reg=self._l2_reg,
labels=self._get_input('labels'),
train=train, scope="PointwiseGeCE", reuse=(not train)
),
train=train
)
def _build_shared_graph(self):
with tf.variable_scope(self._scope, reuse=self._reuse):
self._ge = PointwiseGeCE(
user=self._user_ge,
item=self._item_ge,
item_bias=0,
labels=self._labels,
l2_reg=self._l2_reg,
train=self._train,
scope="ge",
reuse=self._reuse,
)
self._mlp = PointwiseMLPCE(
user=self._user_mlp,
item=self._item_mlp,
item_bias=0,
labels=self._labels,
l2_reg=self._l2_reg,
dims=self._dims,
train=self._train,
dropout=self._dropout,
scope="mlp",
reuse=self._reuse,
)
def _build_default_interactions(self, train=True):
self._add_module(
"interaction",
PointwiseGeCE(
user=self._get_module("user_vec", train=train).get_outputs()[0],
item=self._get_module("item_vec", train=train).get_outputs()[0],
item_bias=self._get_module("item_bias", train=train).get_outputs()[0],
l2_reg=self._l2_reg,
labels=self._get_input("labels"),
train=train,
scope="PointwiseGeCE",
reuse=(not train),
),
train=train,
)
class PointwiseGeMLPCE(Interaction):
"""
The PointwiseGeMLPCE module minimizes the cross entropy classification loss. The logits are calculated as follows [ncf]_ \
(Bias term is not included).
.. math::
\\alpha h^T(u_i^{ge} \odot v_j^{ge}) + (1 - \\alpha)MLP([u_i^{mlp}, v_j^{mlp}])
Parameters
----------
user_mlp: Tensorflow tensor
:math:`u^{mlp}` for users involved in the interactions. Shape: [number of interactions, dimensionality of \
:math:`u^{mlp}`].
user_ge: Tensorflow tensor
:math:`u^{ge}` for users involved in the interactions. Shape: [number of interactions, dimensionality of \
:math:`u^{ge}`].
item_mlp: Tensorflow tensor
:math:`v^{mlp}` for items involved in the interactions. Shape: [number of interactions, dimensionality of \
:math:`v^{mlp}`].
item_ge: Tensorflow tensor
:math:`v^{ge}` for items involved in the interactions. Shape: [number of interactions, dimensionality of \
:math:`v^{ge}`].
item_bias: Tensorflow tensor
Biases for items involved in the interactions. Shape: [number of interactions, 1].
dims: Numpy array.
Specify the size of the MLP (openrec.legacy.modules.extractions.MultiLayerFC).
labels: Tensorflow tensor, required for training.
Groundtruth labels for the interactions. Shape [number of interactions, ].
dropout: float, optional.
Dropout rate for MLP (intermediate layers only).
alpha: float, optional.
Value of :math:`\\alpha`. Default to 0.5.
l2_reg: float, optional
Weight for L2 regularization, i.e., weight decay.
train: bool, optionl
An indicator for training or servining phase.
scope: str, optional
Scope for module variables.
reuse: bool, optional
Whether or not to reuse module variables.
"""
def __init__(
self,
user_mlp,
user_ge,
item_mlp,
item_ge,
item_bias,
dims,
labels=None,
dropout=None,
alpha=0.5,
l2_reg=None,
train=None,
scope=None,
reuse=False,
):
self._user_mlp = user_mlp
self._user_ge = user_ge
self._item_mlp = item_mlp
self._item_ge = item_ge
self._item_bias = item_bias
self._dims = dims
self._alpha = alpha
self._train = train
self._labels = labels
self._dropout = dropout
super(PointwiseGeMLPCE, self).__init__(train=train,
l2_reg=l2_reg,
scope=scope,
reuse=reuse)
def _build_shared_graph(self):
with tf.variable_scope(self._scope, reuse=self._reuse):
self._ge = PointwiseGeCE(
user=self._user_ge,
item=self._item_ge,
item_bias=0,
labels=self._labels,
l2_reg=self._l2_reg,
train=self._train,
scope="ge",
reuse=self._reuse,
)
self._mlp = PointwiseMLPCE(
user=self._user_mlp,
item=self._item_mlp,
item_bias=0,
labels=self._labels,
l2_reg=self._l2_reg,
dims=self._dims,
train=self._train,
dropout=self._dropout,
scope="mlp",
reuse=self._reuse,
)
def _build_training_graph(self):
with tf.variable_scope(self._scope, reuse=self._reuse):
logits = (self._alpha * self._ge.get_outputs()[0] +
(1 - self._alpha) * self._mlp.get_outputs()[0] +
self._item_bias)
labels_float = tf.reshape(tf.to_float(self._labels), (-1, 1))
self._loss = tf.reduce_sum(
tf.nn.sigmoid_cross_entropy_with_logits(labels=labels_float,
logits=logits))
self._outputs.append(logits)
def _build_serving_graph(self):
with tf.variable_scope(self._scope, reuse=self._reuse):
item_bias_rep = tf.tile(self._item_bias,
(tf.shape(self._user)[0], 1))
logits = (
self._alpha * self._ge.get_outputs()[0] +
(1 - self._alpha) * self._mlp.get_outputs()[0] +
tf.reshape(item_bias_rep,
(tf.shape(self._user)[0], tf.shape(self._item)[0])))
self._outputs.append(logits)