File size: 6,221 Bytes
fa98c11 2647d97 fa98c11 2647d97 fa98c11 2647d97 fa98c11 2647d97 fa98c11 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 | """
Script: train_deep.py
Trains the Two-Tower neural recommender and pre-computes item embeddings.
Outputs:
models/two_tower.pt (best checkpoint)
models/item_embeddings.npy (pre-computed item tower outputs)
models/item_index.pkl (item_idx → item metadata lookup)
data/outputs/figures/training_curve.png
Usage:
python scripts/train_deep.py
python scripts/train_deep.py --debug
python scripts/train_deep.py --device cuda
"""
from __future__ import annotations
import argparse
import pickle
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
from torch.utils.data import DataLoader
from src.config import FIGURES_DIR, MODELS_DIR, ensure_dirs, get_config
from src.data.dataset import BPRDataset
from src.models.deep import TwoTowerModel, TwoTowerTrainer, build_item_feature_matrix
from src.utils.logging import get_logger
from src.utils.seed import set_seed
logger = get_logger(__name__)
# Metadata features passed to the item tower
ITEM_META_COLS = [
"item_avg_rating", "item_num_ratings", "item_popularity",
"item_positive_rate", "price_norm", "category_idx", "brand_idx",
]
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Train Two-Tower neural recommender.")
parser.add_argument("--debug", action="store_true")
parser.add_argument("--device", type=str, default=None,
help="Device: 'cpu', 'cuda', 'mps'. Auto-detected if not set.")
return parser.parse_args()
def detect_device(override: str | None) -> str:
if override:
return override
if torch.cuda.is_available():
return "cuda"
if torch.backends.mps.is_available():
return "mps"
return "cpu"
def plot_training_curve(history: list[dict], output_path: Path) -> None:
epochs = [h["epoch"] for h in history]
train_loss = [h["train_loss"] for h in history]
val_loss = [h["val_loss"] for h in history]
fig, ax = plt.subplots(figsize=(8, 4))
ax.plot(epochs, train_loss, label="Train loss", linewidth=2, color="#3b82f6")
ax.plot(epochs, val_loss, label="Val loss", linewidth=2, color="#f97316", linestyle="--")
ax.set_xlabel("Epoch")
ax.set_ylabel("BPR Loss")
ax.set_title("Two-Tower Training Curve (BPR)")
ax.legend()
plt.tight_layout()
output_path.parent.mkdir(parents=True, exist_ok=True)
fig.savefig(output_path, dpi=150)
plt.close(fig)
logger.info("Training curve saved → %s", output_path)
def main() -> None:
args = parse_args()
cfg = get_config(debug=args.debug)
ensure_dirs(cfg)
set_seed(cfg.data.random_seed)
device = detect_device(args.device)
logger.info("Using device: %s", device)
logger.info("Loading data…")
train_df = pd.read_parquet(cfg.data.train_path)
val_df = pd.read_parquet(cfg.data.val_path)
item_features = pd.read_parquet(cfg.data.item_features_path)
# Ensure label column
if "label" not in train_df.columns:
train_df["label"] = (train_df["rating"] >= cfg.data.positive_rating_threshold).astype(int)
if "label" not in val_df.columns:
val_df["label"] = (val_df["rating"] >= cfg.data.positive_rating_threshold).astype(int)
n_users = int(train_df["user_idx"].max() + 1)
n_items = int(item_features["item_idx"].max() + 1)
logger.info("Dataset: %d users, %d items", n_users, n_items)
# Build item metadata feature matrix
item_feat_matrix = build_item_feature_matrix(item_features, ITEM_META_COLS)
n_meta = item_feat_matrix.shape[1]
logger.info("Item metadata features: %d", n_meta)
# BPR Datasets: pairs each positive with a sampled negative
# train_df has pre-sampled negatives (label=0); val_df may only have positives
train_ds = BPRDataset(train_df, item_feat_matrix)
val_ds = BPRDataset(val_df, item_feat_matrix, n_items=n_items)
logger.info("BPR dataset — train pairs: %d, val pairs: %d", len(train_ds), len(val_ds))
train_loader = DataLoader(
train_ds,
batch_size=cfg.deep.batch_size,
shuffle=True,
num_workers=0,
pin_memory=(device == "cuda"),
)
val_loader = DataLoader(
val_ds,
batch_size=cfg.deep.batch_size,
shuffle=False,
num_workers=0,
)
logger.info("Building Two-Tower model…")
model = TwoTowerModel(
n_users=n_users,
n_items=n_items,
embedding_dim=cfg.deep.embedding_dim,
output_dim=cfg.deep.embedding_dim,
hidden_dims=cfg.deep.hidden_dims,
dropout=cfg.deep.dropout,
n_meta_features=n_meta,
)
total_params = sum(p.numel() for p in model.parameters())
logger.info("Model parameters: %d", total_params)
trainer = TwoTowerTrainer(
model=model,
device=device,
lr=cfg.deep.learning_rate,
weight_decay=cfg.deep.weight_decay,
patience=cfg.deep.patience,
model_path=cfg.deep.model_path,
)
logger.info("Training for up to %d epochs…", cfg.deep.epochs)
history = trainer.fit(train_loader, val_loader, epochs=cfg.deep.epochs)
# Plot training curve
plot_training_curve(history, FIGURES_DIR / "training_curve.png")
# Pre-compute and save item embeddings
logger.info("Pre-computing item embeddings…")
item_embeddings = trainer.build_item_embeddings(
n_items=n_items,
item_features=item_feat_matrix,
batch_size=2048,
)
np.save(cfg.deep.item_embeddings_path, item_embeddings)
logger.info("Item embeddings saved → %s shape: %s",
cfg.deep.item_embeddings_path, item_embeddings.shape)
# Save item index (item_idx → item_id + metadata) for explanation lookups
item_index = item_features.set_index("item_idx")[["title", "category", "brand", "price"]].to_dict(orient="index")
with open(cfg.deep.item_index_path, "wb") as f:
pickle.dump(item_index, f)
logger.info("Item index saved → %s", cfg.deep.item_index_path)
logger.info("Two-Tower training complete.")
if __name__ == "__main__":
main()
|