A plain object modeling a batch of `Nodes` or `Edges`.

Do the computation `y = xW + B`. x: [-1, input_dim] W: [intput_dim, output_dim] B: [output_dim] y: [-1, output_dim] Args: name

Encodes the input features of `Data` using `FeatureSpec`. For efficiency, we group the features into `FeatureGroup` accroding to the `FeatureSpec

Provides utility functions for filenames. FileInfo provides easy access to the components of a file's path relative to the project root.

Describes how to encode the values of `Nodes` or `Edges`. Args: feature_num (int, Required): The total feature count. weighted (boolean, O

EmbeddingColumn with dynamic bucket_size.

Uses embedding_lookup to embed the categorical features. Args: name: A unique string identifying the input feature. bucket_size: The size o

Represents a group of `FeatureColumn`s.

GraphSAGE convolution. Args: in_dim: A int indicates dimension of input sample or a list corresponds to dimensional sizes of src and dst

A BatchGraph object, which represents a batch of `SubGraph`s. Nodes, edges in subgraphs are concatenated together and their offsets are recorded

A :code:`Client` class manages a set of remote server to connect and do a round-robin sample between those GraphLearn clients.

SparseEmbeddingColumn with dynamic bucket_size.

EgoGraph based implementation of RGCN. https://arxiv.org/abs/1703.06103. Args: name: A string, layer name. in_dim: An integer or a two ele

FakeNode is used for adding corresponding DagNode of E.outV()/inV() to Dag. E.outV()/inV() doesn't raise any operator, but only changes the field

A HeteroBatchGraph object, which represents a batch of `HeteroSubGraph`s. Each type of Nodes, edges in subgraphs are concatenated together and their

Class for local training and evaluation Args: ckpt_dir: checkpoint dir. ckpt_freq: checkpoint frequency. ckpt_steps: checkpoint steps.

As returned object of `get_next` api of `node_sampler` and `negative_sampler`, as returned object of `get_nodes` of `Graph` or as in-memory objec

Represents real valued or numerical features. Args: name: A unique string identifying the input feature. normalizer_func: If not `None`, a

The structure of io record batch. The record batch is the basic payload of dataloader messages and actor messages. The underlying buffer is cppkafka

Class to manage a GraphLearn server, including start, init and stop.

Uses sparse_embedding_lookup to embed the multivalent categorical feature which is split with delimiter. Args: name: A unique string identif

Induces the edge traversal and it's 1-hop query to SubGraph and generates the additional structure labels.

As returned object of `get_next` api of `edge_sampler` , as returned object of `get_edges` of `Graph` or as in-memory object for constructing gra

`EgoGraph` is a basic data structure used to describe a sampled graph. It constists of src `Data` and src's neighbors(nodes and edges) `Data`. T

Fuses the input feature with the same dimension setting and then lookups embeddings. Args: name: A unique string identifying the input featur

Layer is 1 hop neighbor nodes and the between edges.

As returned object of `get_next` api of `meta_path_sampler`.

SparseNodes is the returned value of full neighbor sampler which is 2D. It can be easily transformed to Tensorflow or PyTorch Sparse Tensors.

`SubGraph` is a basic data structure used to describe a sampled subgraph. It constists of `edge_index` and nodes `Data` and edges `Data`. Args:

Holds 4 copies of all lines with different preprocessing applied to them. 1) elided member contains lines without strings and comments. 2) lines

Custom DistTrainer save node embedding on local file.

Induces the edge traversal and it's 1-hop query to SubGraph.

multi-head GAT convolutional layer.

Handles heterogeneous subgraph(`HeteroSubGraph`) convolution. This layer will perform the convolution operation according to the specified ed

Induces the edge traversal and it's 1-hop query to HeteroSubGraph.

A data object describing heterogeneous `SubGraph`. Different types of nodes and edges are represented by a dict. Args: edge_index_dict: A di

Holds states related to parsing braces.