graph LR
DeepFM["DeepFM"]
_init_graph["_init_graph"]
_initialize_weights["_initialize_weights"]
fit["fit"]
fit_on_batch["fit_on_batch"]
predict["predict"]
evaluate["evaluate"]
get_batch["get_batch"]
DeepFM -- "initializes" --> _init_graph
DeepFM -- "initializes" --> _initialize_weights
DeepFM -- "delegates training control to" --> fit
DeepFM -- "exposes" --> predict
DeepFM -- "exposes" --> evaluate
_init_graph -- "relies on" --> _initialize_weights
_initialize_weights -- "provides initialized weights to" --> _init_graph
fit -- "requests data batches from" --> get_batch
fit -- "dispatches batch-wise training to" --> fit_on_batch
fit -- "invokes" --> evaluate
fit_on_batch -- "receives data from" --> fit
fit_on_batch -- "interacts with" --> _init_graph
predict -- "retrieves input data from" --> get_batch
predict -- "executes forward pass on" --> _init_graph
evaluate -- "leverages" --> predict
evaluate -- "retrieves evaluation data from" --> get_batch
get_batch -- "provides data to" --> fit
get_batch -- "provides data to" --> predict
get_batch -- "provides data to" --> evaluate
The DeepFM Model Core subsystem is primarily defined by the DeepFM class and its associated methods within the DeepFM.py file. This subsystem encapsulates the entire lifecycle of the DeepFM model, from its architectural definition and weight initialization to training, prediction, and evaluation.
Acts as the primary interface for the DeepFM model, orchestrating its entire lifecycle from initialization to training, prediction, and evaluation. It manages the overall flow and coordination of the model's operations.
Related Classes/Methods:
Defines and constructs the TensorFlow computational graph for the DeepFM model, integrating its Factorization Machine (FM) and Deep Neural Network (DNN) components.
Related Classes/Methods:
Sets up and initializes all trainable parameters (weights and biases) required by the DeepFM model's computational graph.
Related Classes/Methods:
Oversees the entire training process, including iterating over epochs, managing data batches, and coordinating batch-wise training and periodic evaluation.
Related Classes/Methods:
Executes a single forward and backward pass on a given data batch, updating model weights based on the computed loss.
Related Classes/Methods:
Generates output predictions for new input data using the trained model.
Related Classes/Methods:
Calculates and reports performance metrics of the model on a given dataset, often by leveraging the predict functionality.
Related Classes/Methods:
Handles the preparation and delivery of data in mini-batches for training, prediction, or evaluation.
Related Classes/Methods: