[medgan-pr-prep] T5: Update generation example (generate_synthetic_mimic3_medgan.py)

[jalengg]

Goal

Rewrite examples/generate_synthetic_mimic3_medgan.py to use the PyHealth 2.0 API (load a trained model, generate samples, save to CSV).

Template

"""Generate synthetic EHR data using a pre-trained MedGAN model."""
import csv
from pyhealth.datasets import MIMIC3Dataset
from pyhealth.tasks import medgan_generation_mimic3_fn
from pyhealth.models import MedGAN

root = "/srv/local/data/physionet.org/files/mimiciii/1.4"
checkpoint_path = "./medgan_save/medgan_checkpoint.pt"
output_path = "./medgan_synthetic_1000.csv"

dataset = MIMIC3Dataset(root=root, tables=["diagnoses_icd", "admissions"])
sample_dataset = dataset.set_task(medgan_generation_mimic3_fn)

model = MedGAN.load_model(checkpoint_path, dataset=sample_dataset)

synthetic = model.synthesize_dataset(num_samples=1000)

with open(output_path, "w", newline="") as f:
    writer = csv.writer(f)
    writer.writerow(["patient_id", "visits"])
    for record in synthetic:
        writer.writerow([record["patient_id"], "|".join(record["visits"])])

print(f"Saved {len(synthetic)} synthetic patients to {output_path}")

Acceptance criteria

• File is examples/generate_synthetic_mimic3_medgan.py
• Uses only PyHealth 2.0 API
• Output is CSV with patient_id and visits columns
• No numpy/binary matrix logic