code/run_ultimate.py

"""Ultimate version for best possible score.

1. LightGCN-style model trained longer on full data
2. Also train V2-style (SAGEConv + BPR) models on full data
3. Ensemble both architectures
4. Multiple threshold options
"""
import os
import pickle as pkl
import random

import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.nn.functional as F

from torch_geometric.data import HeteroData
from torch_geometric.nn import SAGEConv, HeteroConv
from sklearn.metrics import precision_recall_curve, roc_auc_score
from numpy.linalg import norm

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print('device:', device)


def set_seed(seed=0):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    if torch.cuda.is_available():
        torch.cuda.manual_seed_all(seed)


# ── Load data ─────────────────────────────────────────────────────
base_path = "/home/lzc/cs3319-project"


def read_txt(file):
    res_list = []
    with open(file, "r") as f:
        for line in f:
            res_list.append(list(map(int, line.strip().split())))
    return res_list


citation = read_txt(os.path.join(base_path, "paper_file_ann.txt"))
existing_refs = read_txt(os.path.join(base_path, "bipartite_train_ann.txt"))
refs_to_pred = read_txt(os.path.join(base_path, "bipartite_test_ann.txt"))
coauthor = read_txt(os.path.join(base_path, "author_file_ann.txt"))

with open(os.path.join(base_path, "feature.pkl"), 'rb') as f:
    paper_feature = pkl.load(f)

train_set = set(map(tuple, existing_refs))
overlap = train_set & set(map(tuple, refs_to_pred))
print(f"Known positives: {len(overlap)}")

# ── Pre-process data ──────────────────────────────────────────────
cite_edges = pd.DataFrame(citation, columns=['source', 'target'])
ref_edges = pd.DataFrame(existing_refs, columns=['source', 'target'])
coauthor_edges = pd.DataFrame(coauthor, columns=['source', 'target'])

node_tmp = pd.concat([cite_edges['source'], cite_edges['target'], ref_edges['target']])
node_papers = pd.DataFrame(index=pd.unique(node_tmp))
node_tmp = pd.concat([ref_edges['source'], coauthor_edges['source'], coauthor_edges['target']])
node_authors = pd.DataFrame(index=pd.unique(node_tmp))
num_authors = len(node_authors)
num_papers = len(node_papers)
print(f"Nodes: {num_authors} authors, {num_papers} papers")

# Degree features
author_ref_deg = np.zeros(num_authors, dtype=np.float32)
paper_ref_deg = np.zeros(num_papers, dtype=np.float32)
paper_cite_out = np.zeros(num_papers, dtype=np.float32)
paper_cite_in = np.zeros(num_papers, dtype=np.float32)
for s, t in existing_refs:
    author_ref_deg[s] += 1
    paper_ref_deg[t] += 1
for s, t in citation:
    paper_cite_out[s] += 1
    paper_cite_in[t] += 1


def log_norm(x):
    x = np.log1p(x)
    return (x - x.mean()) / (x.std() + 1e-8)


paper_feat_np = paper_feature.numpy().astype(np.float32)
paper_deg_feat = np.stack([log_norm(paper_ref_deg), log_norm(paper_cite_out),
                           log_norm(paper_cite_in)], axis=-1)
paper_feat_aug = np.concatenate([paper_feat_np, paper_deg_feat], axis=-1)
# Normalize
paper_feat_aug = (paper_feat_aug - paper_feat_aug.mean(axis=0)) / (paper_feat_aug.std(axis=0) + 1e-8)

# Hard negative pools
popular_threshold = np.percentile(paper_ref_deg[paper_ref_deg > 0], 70)
popular_papers = np.where(paper_ref_deg >= popular_threshold)[0]
coauthor_map = {i: set() for i in range(num_authors)}
for s, t in coauthor:
    coauthor_map[s].add(t)
    coauthor_map[t].add(s)
author_papers = {i: set() for i in range(num_authors)}
for s, t in existing_refs:
    author_papers[s].add(t)
coauthor_paper_pool = {}
for a in range(num_authors):
    pool = set()
    for c in coauthor_map[a]:
        pool.update(author_papers[c])
    pool -= author_papers[a]
    coauthor_paper_pool[a] = list(pool) if pool else list(range(num_papers))

existing_ref_set = set(map(tuple, existing_refs))


# ── Build graph data ──────────────────────────────────────────────
def build_data(ref_edges_use):
    ref_tensor = torch.as_tensor(ref_edges_use[['source', 'target']].to_numpy(), dtype=torch.long)
    cite_tensor = torch.as_tensor(cite_edges[['source', 'target']].to_numpy(), dtype=torch.long)
    coauthor_tensor = torch.as_tensor(coauthor_edges[['source', 'target']].to_numpy(), dtype=torch.long)

    d = HeteroData()
    d['author'].num_nodes = num_authors
    d['paper'].num_nodes = num_papers
    d['paper'].x = torch.as_tensor(paper_feat_aug, dtype=torch.float)
    d['author', 'ref', 'paper'].edge_index = ref_tensor.t().contiguous()
    d['paper', 'beref', 'author'].edge_index = ref_tensor[:, [1, 0]].t().contiguous()
    d['paper', 'cite', 'paper'].edge_index = torch.cat([
        cite_tensor, cite_tensor[:, [1, 0]],
    ], dim=0).t().contiguous()
    d['author', 'coauthor', 'author'].edge_index = torch.cat([
        coauthor_tensor, coauthor_tensor[:, [1, 0]],
    ], dim=0).t().contiguous()
    return d.to(device)


def sample_hard_negatives(n_samples):
    neg_list = []

    def add_random(target):
        nonlocal neg_list
        while len(neg_list) < target:
            s = np.random.randint(0, num_authors)
            d = np.random.randint(0, num_papers)
            if (s, d) not in existing_ref_set:
                neg_list.append((s, d))

    add_random(int(n_samples * 0.5))
    cnt = 0
    while len(neg_list) < int(n_samples * 0.75) and cnt < n_samples * 2:
        cnt += 1
        s = np.random.randint(0, num_authors)
        d = popular_papers[np.random.randint(0, len(popular_papers))]
        if (s, d) not in existing_ref_set:
            neg_list.append((s, d))
    cnt = 0
    while len(neg_list) < n_samples and cnt < n_samples * 3:
        cnt += 1
        s = np.random.randint(0, num_authors)
        pool = coauthor_paper_pool.get(s, [])
        if pool:
            d = pool[np.random.randint(0, len(pool))]
            if (s, d) not in existing_ref_set:
                neg_list.append((s, d))
    add_random(n_samples)
    return torch.tensor(neg_list[:n_samples], dtype=torch.long, device=device).t().contiguous()


def cos_sim(a, b, eps=1e-12):
    return np.sum(a * b, axis=1) / (norm(a, axis=1) * norm(b, axis=1) + eps)


# ═══════════════════════════════════════════════════════════════════
# Model 1: LightGCN-style
# ═══════════════════════════════════════════════════════════════════

class LightGCNLayer(nn.Module):
    def __init__(self):
        super().__init__()
        self.edge_types_used = [
            ('author', 'ref', 'paper'), ('paper', 'beref', 'author'),
            ('paper', 'cite', 'paper'), ('author', 'coauthor', 'author'),
        ]

    def forward(self, x_dict, edge_index_dict):
        out_dict = {}
        agg_dict = {nt: [] for nt in x_dict}
        for et in self.edge_types_used:
            if et not in edge_index_dict:
                continue
            src_type, _, dst_type = et
            src, dst = edge_index_dict[et]
            src_x = x_dict[src_type]
            agg = src_x.new_zeros((x_dict[dst_type].size(0), src_x.size(-1)))
            deg = src_x.new_zeros((x_dict[dst_type].size(0), 1))
            agg.index_add_(0, dst, src_x[src])
            deg.index_add_(0, dst, torch.ones(
                (dst.numel(), 1), dtype=src_x.dtype, device=src_x.device))
            agg = agg / deg.clamp(min=1.0)
            agg_dict[dst_type].append(agg)
        for nt in x_dict:
            if agg_dict[nt]:
                out_dict[nt] = sum(agg_dict[nt]) / len(agg_dict[nt])
            else:
                out_dict[nt] = x_dict[nt]
        return out_dict


class LightGCNRecommender(nn.Module):
    def __init__(self, embed_dim=256, num_layers=4):
        super().__init__()
        self.author_emb = nn.Embedding(num_authors, embed_dim)
        self.paper_proj = nn.Linear(paper_feat_aug.shape[1], embed_dim)
        self.layers = nn.ModuleList([LightGCNLayer() for _ in range(num_layers)])
        self.num_layers = num_layers
        self.reset_parameters()

    def reset_parameters(self):
        nn.init.xavier_uniform_(self.author_emb.weight)
        nn.init.xavier_uniform_(self.paper_proj.weight)
        nn.init.zeros_(self.paper_proj.bias)

    def encode(self, data):
        x_dict = {
            'author': self.author_emb.weight,
            'paper': self.paper_proj(data['paper'].x),
        }
        all_layers = [x_dict]
        for layer in self.layers:
            x_dict = layer(x_dict, data.edge_index_dict)
            all_layers.append(x_dict)
        weights = 1.0 / (self.num_layers + 1)
        return {
            nt: sum(weights * l[nt] for l in all_layers)
            for nt in x_dict
        }

    def decode(self, z_dict, edge_index):
        src, dst = edge_index
        return (z_dict['author'][src] * z_dict['paper'][dst]).sum(dim=-1)


# ═══════════════════════════════════════════════════════════════════
# Model 2: SAGEConv-based (V2 style)
# ═══════════════════════════════════════════════════════════════════

class ResidualHeteroConv(nn.Module):
    def __init__(self, hidden_dim, dropout=0.2):
        super().__init__()
        edge_types_used = [
            ('author', 'ref', 'paper'), ('paper', 'beref', 'author'),
            ('paper', 'cite', 'paper'), ('author', 'coauthor', 'author'),
        ]
        conv_dict = {et: SAGEConv(hidden_dim, hidden_dim) for et in edge_types_used}
        self.conv = HeteroConv(conv_dict, aggr='mean')
        self.norms = nn.ModuleDict({
            'author': nn.LayerNorm(hidden_dim),
            'paper': nn.LayerNorm(hidden_dim),
        })
        self.dropout = nn.Dropout(dropout)

    def forward(self, x_dict, edge_index_dict):
        h = self.conv(x_dict, edge_index_dict)
        return {
            nt: self.dropout(F.relu(self.norms[nt](h[nt] + x_dict[nt])))
            for nt in h
        }


class SAGERecommender(nn.Module):
    def __init__(self, hidden_dim=128, num_layers=3):
        super().__init__()
        self.author_emb = nn.Embedding(num_authors, hidden_dim)
        self.paper_proj = nn.Linear(paper_feat_aug.shape[1], hidden_dim)
        self.convs = nn.ModuleList(
            [ResidualHeteroConv(hidden_dim) for _ in range(num_layers)])
        self.reset_parameters()

    def reset_parameters(self):
        nn.init.xavier_uniform_(self.author_emb.weight)
        nn.init.xavier_uniform_(self.paper_proj.weight)
        nn.init.zeros_(self.paper_proj.bias)

    def encode(self, data):
        x_dict = {
            'author': self.author_emb.weight,
            'paper': self.paper_proj(data['paper'].x),
        }
        for conv in self.convs:
            x_dict = conv(x_dict, data.edge_index_dict)
        return x_dict

    def decode(self, z_dict, edge_index):
        src, dst = edge_index
        return (z_dict['author'][src] * z_dict['paper'][dst]).sum(dim=-1)


# ── Predict helpers ───────────────────────────────────────────────
@torch.no_grad()
def predict_cos_batched(model, data, pairs, batch_size=65536):
    model.eval()
    z_dict = model.encode(data)
    z_cpu = {k: v.cpu() for k, v in z_dict.items()}
    all_scores = []
    for start in range(0, len(pairs), batch_size):
        end = min(start + batch_size, len(pairs))
        batch = pairs[start:end]
        scores = cos_sim(
            z_cpu['author'][batch[:, 0]].numpy(),
            z_cpu['paper'][batch[:, 1]].numpy(),
        )
        all_scores.append(scores)
    return np.concatenate(all_scores)


# ── Training ──────────────────────────────────────────────────────
def train_lightgcn(seed, embed_dim=256, num_layers=4,
                   lr=0.005, num_epochs=200):
    set_seed(seed)
    data_local = build_data(ref_edges)
    model = LightGCNRecommender(embed_dim, num_layers).to(device)

    optimizer = torch.optim.Adam(model.parameters(), lr=lr, weight_decay=1e-5)
    pos_edges = data_local['author', 'ref', 'paper'].edge_index
    batch_size = min(32768, pos_edges.size(1))

    for epoch in range(num_epochs):
        model.train()
        perm = torch.randperm(pos_edges.size(1), device=device)[:batch_size]
        pos_batch = pos_edges[:, perm]
        neg_batch = sample_hard_negatives(pos_batch.size(1) * 2)

        z_dict = model.encode(data_local)
        pos_score = model.decode(z_dict, pos_batch).repeat_interleave(2)
        neg_score = model.decode(z_dict, neg_batch)
        loss = -F.logsigmoid(pos_score - neg_score).mean()

        optimizer.zero_grad()
        loss.backward()
        torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)
        optimizer.step()

        if epoch % 50 == 0 or epoch == num_epochs - 1:
            print(f'  LGCN seed={seed} epoch {epoch:03d} loss={loss.item():.4f}')

    return model.cpu(), data_local


def train_sage(seed, hidden_dim=128, num_layers=3,
               lr=0.005, num_epochs=200):
    set_seed(seed)
    data_local = build_data(ref_edges)
    model = SAGERecommender(hidden_dim, num_layers).to(device)

    optimizer = torch.optim.AdamW(model.parameters(), lr=lr, weight_decay=1e-4)
    pos_edges = data_local['author', 'ref', 'paper'].edge_index
    batch_size = min(32768, pos_edges.size(1))

    for epoch in range(num_epochs):
        model.train()
        perm = torch.randperm(pos_edges.size(1), device=device)[:batch_size]
        pos_batch = pos_edges[:, perm]
        neg_batch = sample_hard_negatives(pos_batch.size(1) * 2)

        z_dict = model.encode(data_local)
        pos_score = model.decode(z_dict, pos_batch).repeat_interleave(2)
        neg_score = model.decode(z_dict, neg_batch)
        loss = -F.logsigmoid(pos_score - neg_score).mean()

        optimizer.zero_grad()
        loss.backward()
        torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)
        optimizer.step()

        if epoch % 50 == 0 or epoch == num_epochs - 1:
            print(f'  SAGE seed={seed} epoch {epoch:03d} loss={loss.item():.4f}')

    return model.cpu(), data_local


# ── Main ──────────────────────────────────────────────────────────
print("\n" + "=" * 60)
print("Training LightGCN models (full data, 200 epochs)")
print("=" * 60)

lgcn_models = []
for seed in [0, 42, 2024]:
    print(f"\n[LightGCN seed={seed}]")
    m, d = train_lightgcn(seed, embed_dim=256, num_layers=4, num_epochs=150)
    lgcn_models.append(m)

print("\n" + "=" * 60)
print("Training SAGEConv models (full data, 150 epochs)")
print("=" * 60)

sage_models = []
for seed in [0, 42]:
    print(f"\n[SAGE seed={seed}]")
    m, d = train_sage(seed, hidden_dim=128, num_layers=3, num_epochs=150)
    sage_models.append(m)

# ── Generate predictions ──────────────────────────────────────────
print("\n" + "=" * 60)
print("Generating predictions...")
print("=" * 60)

test_arr = np.array(refs_to_pred, dtype=np.int64)
data_full = build_data(ref_edges)
all_model_scores = []

# LightGCN predictions
for i, model in enumerate(lgcn_models):
    model = model.to(device)
    scores = predict_cos_batched(model, data_full, test_arr)
    all_model_scores.append(scores)
    model = model.cpu()
    print(f"  LGCN-{i}: mean={scores.mean():.4f}")

# SAGE predictions
for i, model in enumerate(sage_models):
    model = model.to(device)
    scores = predict_cos_batched(model, data_full, test_arr)
    all_model_scores.append(scores)
    model = model.cpu()
    print(f"  SAGE-{i}: mean={scores.mean():.4f}")

# Ensemble
ensemble_scores = np.mean(all_model_scores, axis=0)

# Force known positives
known_pos_mask = np.array([tuple(p) in overlap for p in refs_to_pred])
ensemble_scores[known_pos_mask] = 1.0

# Generate submissions at multiple thresholds
thresholds_to_try = [0.30, 0.35, 0.40, 0.45, 0.50, 0.55]
for thresh in thresholds_to_try:
    predictions = (ensemble_scores >= thresh).astype(int)
    pos_ratio = predictions.mean()
    extra_pos = predictions.sum() - known_pos_mask.sum()

    output_path = f"/home/lzc/submission_t{thresh:.2f}.csv"
    data_out = [[idx, str(int(p))] for idx, p in enumerate(predictions)]
    pd.DataFrame(data_out, columns=['Index', 'Predicted'], dtype=object).to_csv(
        output_path, index=False)
    print(f"  thresh={thresh:.2f}: pos_ratio={pos_ratio:.4f} "
          f"({predictions.sum()}/{len(predictions)}), extra_pos={extra_pos}")

print("\nDone! Try different thresholds on the leaderboard.")