hopwise.model.context_aware_recommender.kd_dagfm¶

Reference:: Zhen Tian et al. “Directed Acyclic Graph Factorization Machines for CTR Prediction via Knowledge Distillation.” in WSDM 2023.
Reference code:: https://github.com/chenyuwuxin/DAGFM

Classes¶

`KD_DAGFM`	KD_DAGFM is a context-based recommendation model. The model is based on directed acyclic graph and knowledge
`DAGFM`
`CrossNet`
`CINComp`
`CIN`

Module Contents¶

class hopwise.model.context_aware_recommender.kd_dagfm.KD_DAGFM(config, dataset)¶

Bases: hopwise.model.abstract_recommender.ContextRecommender

KD_DAGFM is a context-based recommendation model. The model is based on directed acyclic graph and knowledge distillation. It can learn arbitrary feature interactions from the complex teacher networks and achieve approximately lossless model performance. It can also greatly reduce the computational resource costs.

phase¶

alpha¶

beta¶

student_network¶

teacher_network¶

loss_fn¶

get_teacher_config(config)¶

FeatureInteraction(feature)¶

forward(interaction)¶

calculate_loss(interaction)¶

Calculate the training loss for a batch data.

Parameters:: interaction (Interaction) – Interaction class of the batch.
Returns:: Training loss, shape: []
Return type:: torch.Tensor

predict(interaction)¶

Predict the scores between users and items.

Parameters:: interaction (Interaction) – Interaction class of the batch.
Returns:: Predicted scores for given users and items, shape: [batch_size]
Return type:: torch.Tensor

class hopwise.model.context_aware_recommender.kd_dagfm.DAGFM(config)¶

Bases: torch.nn.Module

type¶

depth¶

adj_matrix¶

connect_layer¶

linear¶

FeatureInteraction(feature)¶

class hopwise.model.context_aware_recommender.kd_dagfm.CrossNet(config)¶

Bases: torch.nn.Module

depth¶

embedding_size¶

feature_num¶

in_feature_num¶

cross_layer_w¶

bias¶

linear¶

FeatureInteraction(x_0)¶

forward(feature)¶

class hopwise.model.context_aware_recommender.kd_dagfm.CINComp(indim, outdim, config)¶

Bases: torch.nn.Module

conv¶

forward(feature, base)¶

class hopwise.model.context_aware_recommender.kd_dagfm.CIN(config)¶

Bases: torch.nn.Module

cinlist¶

cin¶

linear¶

backbone = ['cin', 'linear']¶

loss_fn¶

FeatureInteraction(feature)¶

forward(feature)¶