API: the `NEExT` class

NEExT.framework.NEExT is the main entry point, exported as from NEExT import NEExT. Signatures below match the source in NEExT/framework.py (v0.3.1).

Constructor

signature

NEExT(log_level: str = "INFO")

set_log_level(level: str) -> None changes the level later ("DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL").

Loading

read_from_csv(
  edges_path,
  node_graph_mapping_path,
  graph_label_path=None,
  node_features_path=None,
  edge_features_path=None,
  graph_type="networkx",
  reindex_nodes=True,
  filter_largest_component=True,
  node_sample_rate=1.0,
) -> GraphCollection

load_from_networkx(
  nx_graphs,
  graph_type="networkx",
  reindex_nodes=True,
  filter_largest_component=True,
  node_sample_rate=1.0,
) -> GraphCollection

See Data loading. DataFrame loading lives on GraphIO.load_from_dfs.

Features

features

compute_node_features(
  graph_collection,
  feature_list,
  feature_vector_length=3,
  normalize_features=True,
  show_progress=True,
  n_jobs=1,
  my_feature_methods=None,
  parallel_backend="loky",      # "loky" | "threading"
  profile_features=False,
  joblib_kwargs=None,
) -> Features

See Structural features.

Embeddings

embeddings

compute_graph_embeddings(
  graph_collection,
  features,
  embedding_algorithm,          # "approx_wasserstein" | "wasserstein" | "sinkhornvectorizer" | "gnn"
  embedding_dimension,
  feature_columns=None,
  random_state=42,
  memory_size="4G",
  # GNN-only (used when embedding_algorithm == "gnn"):
  architecture="GCN",           # "GCN" | "GraphSAGE" | "GIN"
  hidden_dims=None,             # default [64, 32]
  epochs=100,
  learning_rate=0.01,
  weight_decay=5e-4,
  dropout=0.0,
  pooling="mean",               # "mean" | "sum" | "max"
  early_stopping_patience=10,
) -> Embeddings

See Embeddings.

Models & importance

models

train_ml_model(
  graph_collection,
  embeddings,
  model_type,                   # "classifier" | "regressor"
  balance_dataset=False,
  sample_size=5,
  n_jobs=-1,
  parallel_backend="process",
) -> dict

compute_feature_importance(
  graph_collection,
  features,
  feature_importance_algorithm, # "supervised_greedy" | "supervised_fast" | "unsupervised"
  embedding_algorithm="approx_wasserstein",
  random_state=42,
  n_iterations=5,
  n_jobs=-1,
  parallel_backend="process",
) -> pandas.DataFrame

See ML models and Feature importance.

Egonets

egonets

compute_k_hop_egonets(
  graph_collection,
  k_hop=1,
  egonet_feature_target=None,
  skip_features=None,
  nodes_to_sample=None,
  sample_fraction=1.0,
  random_seed=13,
) -> EgonetCollection

compute_leiden_egonets(           # requires the iGraph backend
  graph_collection,
  egonet_feature_target=None,
  skip_features=None,
  n_iterations=10,
  resolution=1.0,
) -> EgonetCollection

See Egonets.

Synthetic graphs

synthetic

generate_synthetic_graphs(
  preset=None,                  # e.g. "er_vs_ba", "multiclass_topology", ...
  specs=None,                   # list of (GeneratorSpec, count)
  seed=42,
  **kwargs,                     # forwarded to the preset
) -> GraphCollection

See Synthetic graphs.

API: the NEExT class