anomalydae_experiment.py

# Copyright 2021 Grabtaxi Holdings Pte Ltd (GRAB), All rights reserved.
# Use of this source code is governed by an MIT-style license that can be found in the LICENSE file

import sys

from sklearn.metrics import roc_curve, precision_recall_curve, auc

from data_finefoods import load_graph

import argparse
import os

import torch
from torch_geometric.data import Data
from torch_scatter import scatter
from utils.seed import seed_all

# train a detector
from pygod.models import AnomalyDAE

# %% args

parser = argparse.ArgumentParser(description="AnomalyDAE")
parser.add_argument("--name", type=str, default="wikipedia_anomaly", help="name")
parser.add_argument(
    "--key", type=str, default="graph_anomaly_list", help="key to the data"
)
parser.add_argument("--id", type=int, default=0, help="id to the data")
parser.add_argument("--n-epoch", type=int, default=200, help="number of epoch")
parser.add_argument(
    "--num-neighbors", type=int, default=-1, help="number of neighbors for node"
)
parser.add_argument("--batch-size", type=int, default=0, help="batch size")
parser.add_argument("--alpha", type=float, default=0.8, help="balance parameter")
parser.add_argument("--lr", type=float, default=1e-2, help="learning rate")
parser.add_argument("--gpu", type=int, default=0, help="gpu number")

args1 = vars(parser.parse_args())

args2 = {
    "seed": 0,
    "hidden_channels": 32,
    "dropout_prob": 0.0,
}

args = {**args1, **args2}

seed_all(args["seed"])

result_dir = "results/"

# %% data
data = load_graph(args["name"], args["key"], args["id"])

u_ch = data.xu.shape[1]
v_ch = data.xv.shape[1]
e_ch = data.xe.shape[1]

print(
    f"Data dimension: U node = {data.xu.shape}; V node = {data.xv.shape}; E edge = {data.xe.shape}; \n"
)

# %% model

xu, xv = data.xu, data.xv
xe, adj = data.xe, data.adj
yu, yv, ye = data.yu, data.yv, data.ye


# %% to homogen
nu = xu.shape[0]
nv = xv.shape[0]
nn = nu + nv

# to homogen
row_h = torch.cat([adj.storage.row(), adj.storage.col() + nu])
col_h = torch.cat([adj.storage.col() + nu, adj.storage.row()])
edge_index_h = torch.stack([row_h, col_h])
xuh = torch.cat(
    [
        scatter(xe, adj.storage.row(), dim=0, reduce="max"),
        scatter(xe, adj.storage.row(), dim=0, reduce="mean"),
    ],
    dim=1,
)
xvh = torch.cat(
    [
        scatter(xe, adj.storage.col(), dim=0, reduce="max"),
        scatter(xe, adj.storage.col(), dim=0, reduce="mean"),
    ],
    dim=1,
)
xh = torch.cat([xuh, xvh], dim=0)
yh = torch.cat([yu, yv], dim=0)
data_h = Data(x=xh, edge_index=edge_index_h, y=yh)

# %% model

device = torch.device(f'cuda:{args["gpu"]}' if torch.cuda.is_available() else "cpu")
model = AnomalyDAE(
    embed_dim=args["hidden_channels"],
    out_dim=args["hidden_channels"],
    dropout=args["dropout_prob"],
    alpha=args["alpha"],
    epoch=args["n_epoch"],
    lr=args["lr"],
    verbose=True,
    gpu=args["gpu"],
    batch_size=args["batch_size"],
    num_neigh=args["num_neighbors"],
)

print(args)
print()


def auc_eval(pred, y):

    rc_curve = roc_curve(y, pred)
    pr_curve = precision_recall_curve(y, pred)
    roc_auc = auc(rc_curve[0], rc_curve[1])
    pr_auc = auc(pr_curve[1], pr_curve[0])

    return roc_auc, pr_auc, rc_curve, pr_curve


# %% run training

model.fit(data_h, yh)
score = model.decision_scores_

score_u = score[:nu]
score_v = score[nu:]
score_e_u = score_u[adj.storage.row().numpy()]
score_e_v = score_v[adj.storage.col().numpy()]
score_e = (score_e_u + score_e_v) / 2

u_roc_auc, u_pr_auc, u_rc_curve, u_pr_curve = auc_eval(score_u, yu.numpy())
v_roc_auc, v_pr_auc, v_rc_curve, v_pr_curve = auc_eval(score_v, yv.numpy())
e_roc_auc, e_pr_auc, e_rc_curve, e_pr_curve = auc_eval(score_e, ye.numpy())

print(
    f"Eval | "
    + f"u auc-roc: {u_roc_auc:.4f}, v auc-roc: {v_roc_auc:.4f}, e auc-roc: {e_roc_auc:.4f} | "
    + f"u auc-pr {u_pr_auc:.4f}, v auc-pr {v_pr_auc:.4f}, e auc-pr {e_pr_auc:.4f}"
)

auc_metrics = {
    "u_roc_auc": u_roc_auc,
    "u_pr_auc": u_pr_auc,
    "v_roc_auc": v_roc_auc,
    "v_pr_auc": v_pr_auc,
    "e_roc_auc": e_roc_auc,
    "e_pr_auc": e_pr_auc,
    "u_roc_curve": u_rc_curve,
    "u_pr_curve": u_pr_curve,
    "v_roc_curve": v_rc_curve,
    "v_pr_curve": v_pr_curve,
    "e_roc_curve": e_rc_curve,
    "e_pr_curve": e_pr_curve,
}
anomaly_score = {"score_u": score_u, "score_v": score_v, "score_e": score_e}

model_stored = {
    "args": args,
    "auc_metrics": auc_metrics,
    "state_dict": model.model.state_dict(),
}
output_stored = {"args": args, "anomaly_score": anomaly_score}

print("Saving current results...")
torch.save(
    model_stored,
    os.path.join(
        result_dir,
        f"anomalydae-{args['name']}-{args['id']}-alpha-{args['alpha']}-model.th",
    ),
)
torch.save(
    output_stored,
    os.path.join(
        result_dir,
        f"anomalydae-{args['name']}-{args['id']}-alpha-{args['alpha']}-output.th",
    ),
)


print()
print(args)