def direct_neighbors(
adj_matrix_file, train_node_labels_file, test_node_labels_file, use_cuda, metadata
):
mlflow.set_experiment("Predict")
with mlflow.start_run(run_name=RUN_NAME):
u_mlflow.add_metadata(metadata)
mlflow.set_tag("use_cuda", use_cuda)
mlflow.log_param("model", MODEL_NAME)
train_labels = data_loaders.load_labels(
train_node_labels_file, use_cuda=use_cuda
)
test_labels = data_loaders.load_labels(test_node_labels_file, use_cuda=use_cuda)
labels = (train_labels.byte() | test_labels.byte()).long()
n_nodes = labels.size(0)
graph = data_loaders.load_graph(
[adj_matrix_file],
n_nodes,
self_loop=False,
normalization=None,
use_cuda=use_cuda,
)
print(RUN_NAME)
ranks_df = predict.run(labels=labels, model_class=DirectNeighbors, graph=graph)
u_mlflow.log_dataframe(ranks_df, "predictions", "results")
def save_ranks(ranks_df, n_nodes, run_name, params=None):
u_mlflow.log_dataframe(ranks_df, "ranks", "results")
cum_dist = cumulative_distribution_function(
ranks=ranks_df["rank"].values, n_nodes=n_nodes
)
u_mlflow.log_ndarray(cum_dist, "cum_dist", "results")
for index, n in enumerate(cum_dist):
mlflow.log_metric("cum_dist", n, step=index)
# AUC
x = np.arange(1, n_nodes + 1)
auc = metrics.auc(x=x, y=cum_dist)
mlflow.log_metric("auc", auc)
# AUC ratio
mlflow.log_metric("auc_ratio", auc / (n_nodes - 1))
# Median rank
median = np.median(ranks_df["rank"].values)
mlflow.log_metric("median_rank", median)
# Plot
fig = plot_cumulative_distribution(x, cum_dist, auc, run_name, params)
u_mlflow.log_fig(fig, "cumulative_distribution", "figures")
plt.close(fig)
def bagging_mlp(
embeddings_file,
train_node_labels_file,
test_node_labels_file,
node_features_file,
use_cuda,
params,
metadata,
):
mlflow.set_experiment("Predict")
with mlflow.start_run(run_name=RUN_NAME):
mlflow.log_param("model", MODEL_NAME)
u_mlflow.add_params(**params)
u_mlflow.add_metadata(metadata)
mlflow.set_tag("use_cuda", use_cuda)
train_labels = data_loaders.load_labels(train_node_labels_file,
use_cuda=use_cuda)
test_labels = data_loaders.load_labels(test_node_labels_file,
use_cuda=use_cuda)
labels = (train_labels.byte() | test_labels.byte()).long()
embeddings = None
node_features = None
if embeddings_file is not None:
mlflow.log_param("embeddings", True)
mlflow.log_artifact(embeddings_file, "inputs")
embeddings = data_loaders.load_embeddings(embeddings_file,
use_cuda=use_cuda)
else:
mlflow.log_param("embeddings", False)
if node_features_file is not None:
mlflow.log_param("node_features", True)
mlflow.log_artifact(node_features_file, "inputs")
node_features = data_loaders.load_node_features(
node_features_file, use_cuda)
else:
mlflow.log_param("node_features", False)
if embeddings is not None and node_features is not None:
in_features = torch.cat((embeddings, node_features), dim=1)
elif embeddings is not None:
in_features = embeddings
elif node_features is not None:
in_features = node_features
print(RUN_NAME)
ranks_df = predict.run(labels=labels,
model_class=Bagging,
bagging_model=MLP,
features=in_features,
**params)
u_mlflow.log_dataframe(ranks_df, "predictions", "results")
def bagging_rgcn(
adj_matrix_files,
train_node_labels_file,
test_node_labels_file,
node_features_file,
use_cuda,
params,
metadata,
):
mlflow.set_experiment("Predict")
with mlflow.start_run(run_name=RUN_NAME):
mlflow.log_param("model", MODEL_NAME)
u_mlflow.add_params(**params)
u_mlflow.add_metadata(metadata)
mlflow.set_tag("use_cuda", use_cuda)
train_labels = data_loaders.load_labels(train_node_labels_file,
use_cuda=use_cuda)
test_labels = data_loaders.load_labels(test_node_labels_file,
use_cuda=use_cuda)
labels = (train_labels.byte() | test_labels.byte()).long()
n_nodes = labels.size(0)
if node_features_file is not None:
mlflow.log_param("node_features", True)
mlflow.log_artifact(node_features_file, "inputs")
features = data_loaders.load_node_features(node_features_file,
use_cuda)
else:
mlflow.log_param("node_features", False)
features = None
graph = data_loaders.load_graph(
adj_matrix_files,
n_nodes,
add_edge_type=True,
add_node_ids=True,
normalization=NORMALIZATION,
use_cuda=use_cuda,
)
print(RUN_NAME)
ranks_df = predict.run(labels=labels,
model_class=Bagging,
bagging_model=RGCN,
features=features,
graph=graph,
n_rels=len(adj_matrix_files),
**params)
u_mlflow.log_dataframe(ranks_df, "predictions", "results")
def net_prop_with_restart(
run_name,
model_name,
normalization,
adj_matrix_file,
train_node_labels_file,
test_node_labels_file,
use_cuda,
params,
metadata,
):
mlflow.set_experiment("Predict")
with mlflow.start_run(run_name=run_name):
u_mlflow.add_params(**params)
u_mlflow.add_metadata(metadata)
mlflow.log_param("model", model_name)
mlflow.set_tag("use_cuda", use_cuda)
train_labels = data_loaders.load_labels(
train_node_labels_file, use_cuda=use_cuda
)
test_labels = data_loaders.load_labels(test_node_labels_file, use_cuda=use_cuda)
labels = (train_labels.byte() | test_labels.byte()).long()
adjacency_matrix = data_loaders.load_adj_matrices(
[adj_matrix_file], normalization=normalization, use_cuda=use_cuda
)[0]
print(run_name)
ranks_df = predict.run(
labels=labels,
model_class=NetPropWithRestart,
adjacency_matrix=adjacency_matrix,
**params
)
u_mlflow.log_dataframe(ranks_df, "predictions", "results")
def rwr_m(
adj_matrix_files,
train_node_labels_file,
test_node_labels_file,
use_cuda,
params,
metadata,
):
mlflow.set_experiment("Predict")
with mlflow.start_run(run_name=RUN_NAME):
mlflow.log_param("model", MODEL_NAME)
u_mlflow.add_params(**params)
u_mlflow.add_metadata(metadata)
mlflow.set_tag("use_cuda", use_cuda)
train_labels = data_loaders.load_labels(train_node_labels_file,
use_cuda=use_cuda)
test_labels = data_loaders.load_labels(test_node_labels_file,
use_cuda=use_cuda)
labels = (train_labels.byte() | test_labels.byte()).long()
print("Loading adjacency matrices")
adj_matrices = data_loaders.load_adj_matrices(adj_matrix_files,
normalization=None,
use_cuda=use_cuda)
adjacency_matrices = torch.stack(adj_matrices)
print(RUN_NAME)
ranks_df = predict.run(labels=labels,
model_class=RwrM,
adjacency_matrices=adjacency_matrices,
**params)
u_mlflow.log_dataframe(ranks_df, "predictions", "results")
