Inference_&_LLM/inference.py

import gymnasium as gym
import sinergym  # noqa: F401 (registers envs)
import pandas as pd
import numpy as np
import os
import sinergym
import json
import sys
from unihvac.find_files import (
    detect_paths,
    find_manifest,
    find_building_and_weather_from_manifest,
)
from unihvac.tables import (
    print_monthly_tables_extra,
    print_monthly_tables_split,
)
from unihvac.rollout import run_rollout_to_df


# ============================================
# FOR TABLE
pd.set_option("display.max_columns", None)
pd.set_option("display.width", 240)
pd.set_option("display.max_colwidth", 32)
pd.set_option("display.float_format", lambda x: f"{x:,.2f}")
# ============================================

# ==============================================================================
#  USER CONFIGURATION
# ==============================================================================
TARGET_LOCATION = "Atlanta"  # Buffalo, Miami, Dubai, Fairbanks, HoChiMinh
TARGET_THERMAL = "default"   # default, high_performance, low_performance
TARGET_OCCUPANCY = "standard"         # standard, school, retail, etc.

# Baseline-like setpoints (also used as DT seed)
HEATING_SP = 21.0
COOLING_SP = 24.0

# Choose policy mode: "dt" or "rbc"
POLICY_TYPE = "dt"   # change to "rbc" to match baseline runner exactly


# ==========================================
# PATH DISCOVERY (ROBUST)
# ==========================================
paths = detect_paths(outputs_dirname="baseline_results")
manifest_path = find_manifest(paths, building="OfficeSmall", prefer_patched=True)
output_root = str(paths.outputs_root)
os.makedirs(output_root, exist_ok=True)
TIME_STEP_HOURS = 900.0 / 3600.0  # 0.25 h

# ==========================================
# ACTUATORS & VARIABLES (keep identical)
# ==========================================
hot_actuators = {
    "Htg_Core": ("Zone Temperature Control", "Heating Setpoint", "CORE_ZN"),
    "Clg_Core": ("Zone Temperature Control", "Cooling Setpoint", "CORE_ZN"),
    "Htg_P1": ("Zone Temperature Control", "Heating Setpoint", "PERIMETER_ZN_1"),
    "Clg_P1": ("Zone Temperature Control", "Cooling Setpoint", "PERIMETER_ZN_1"),
    "Htg_P2": ("Zone Temperature Control", "Heating Setpoint", "PERIMETER_ZN_2"),
    "Clg_P2": ("Zone Temperature Control", "Cooling Setpoint", "PERIMETER_ZN_2"),
    "Htg_P3": ("Zone Temperature Control", "Heating Setpoint", "PERIMETER_ZN_3"),
    "Clg_P3": ("Zone Temperature Control", "Cooling Setpoint", "PERIMETER_ZN_3"),
    "Htg_P4": ("Zone Temperature Control", "Heating Setpoint", "PERIMETER_ZN_4"),
    "Clg_P4": ("Zone Temperature Control", "Cooling Setpoint", "PERIMETER_ZN_4"),
}

hot_variables = {
    "outdoor_temp": ("Site Outdoor Air DryBulb Temperature", "Environment"),
    "core_temp": ("Zone Air Temperature", "Core_ZN"),
    "perim1_temp": ("Zone Air Temperature", "Perimeter_ZN_1"),
    "perim2_temp": ("Zone Air Temperature", "Perimeter_ZN_2"),
    "perim3_temp": ("Zone Air Temperature", "Perimeter_ZN_3"),
    "perim4_temp": ("Zone Air Temperature", "Perimeter_ZN_4"),
    "elec_power": ("Facility Total HVAC Electricity Demand Rate", "Whole Building"),
    "core_occ_count": ("Zone People Occupant Count", "CORE_ZN"),
    "perim1_occ_count": ("Zone People Occupant Count", "PERIMETER_ZN_1"),
    "perim2_occ_count": ("Zone People Occupant Count", "PERIMETER_ZN_2"),
    "perim3_occ_count": ("Zone People Occupant Count", "PERIMETER_ZN_3"),
    "perim4_occ_count": ("Zone People Occupant Count", "PERIMETER_ZN_4"),

    "outdoor_dewpoint": ("Site Outdoor Air Dewpoint Temperature", "Environment"),
    "outdoor_wetbulb": ("Site Outdoor Air Wetbulb Temperature", "Environment"),

    "core_rh": ("Zone Air Relative Humidity", "CORE_ZN"),
    "perim1_rh": ("Zone Air Relative Humidity", "PERIMETER_ZN_1"),
    "perim2_rh": ("Zone Air Relative Humidity", "PERIMETER_ZN_2"),
    "perim3_rh": ("Zone Air Relative Humidity", "PERIMETER_ZN_3"),
    "perim4_rh": ("Zone Air Relative Humidity", "PERIMETER_ZN_4"),

    "core_ash55_notcomfortable_summer": ("Zone Thermal Comfort ASHRAE 55 Simple Model Summer Clothes Not Comfortable Time", "CORE_ZN"),
    "core_ash55_notcomfortable_winter": ("Zone Thermal Comfort ASHRAE 55 Simple Model Winter Clothes Not Comfortable Time", "CORE_ZN"),
    "core_ash55_notcomfortable_any": ("Zone Thermal Comfort ASHRAE 55 Simple Model Summer or Winter Clothes Not Comfortable Time", "CORE_ZN"),

    "p1_ash55_notcomfortable_any": ("Zone Thermal Comfort ASHRAE 55 Simple Model Summer or Winter Clothes Not Comfortable Time", "PERIMETER_ZN_1"),
    "p2_ash55_notcomfortable_any": ("Zone Thermal Comfort ASHRAE 55 Simple Model Summer or Winter Clothes Not Comfortable Time", "PERIMETER_ZN_2"),
    "p3_ash55_notcomfortable_any": ("Zone Thermal Comfort ASHRAE 55 Simple Model Summer or Winter Clothes Not Comfortable Time", "PERIMETER_ZN_3"),
    "p4_ash55_notcomfortable_any": ("Zone Thermal Comfort ASHRAE 55 Simple Model Summer or Winter Clothes Not Comfortable Time", "PERIMETER_ZN_4"),
}


class BaselineReward:
    def __init__(self, *args, **kwargs):
        pass
    def __call__(self, obs_dict):
        return 0.0, {}


def run_eval_for_location(location, building_path, weather_path):
    print("\n" + "=" * 80)
    print(f"Running eval for location: {location}")
    print(f" Building: {building_path}")
    print(f" Weather: {weather_path}")
    print(f" Policy: {POLICY_TYPE}")
    print("=" * 80)

    out_dir = os.path.join(output_root, location)
    os.makedirs(out_dir, exist_ok=True)

    # Build policy (DT or RBC) — policy state stays outside policy_fn
    if POLICY_TYPE == "dt":
        RUN_DIR = "Trajectories_code/run_007"  # update
        policy = make_policy(
            "dt",
            ckpt_path=os.path.join(RUN_DIR, "ckpt_10.pt"),
            model_config_path=os.path.join(RUN_DIR, "model_config.json"),
            norm_stats_path="Trajectories_code/traj_results/norm_stats.npz",
            context_len=24,
            max_tokens_per_step=64,
        )
    else:
        policy = make_policy("rbc", heating_sp=HEATING_SP, cooling_sp=COOLING_SP)

    policy.reset()

    def policy_fn(obs, info, step):
        if step == 0:
            print("OBS TYPE:", type(obs), "SHAPE:", getattr(obs, "shape", None))
            if isinstance(obs, dict):
                print("OBS KEYS SAMPLE:", list(obs.keys())[:10])
        action, _, _ = policy.act(obs, info, step)
        return action

    df = run_rollout_to_df(
        building_path=str(building_path),
        weather_path=str(weather_path),
        variables=hot_variables,
        actuators=hot_actuators,
        policy_fn=policy_fn,
        location=location,
        timestep_hours=TIME_STEP_HOURS,
        heating_sp=HEATING_SP,   
        cooling_sp=COOLING_SP,
        reward=BaselineReward,
        max_steps=None,
        verbose=True,
    )

    print("setpoint_htg min/max:", df["setpoint_htg"].min(), df["setpoint_htg"].max())
    print("setpoint_clg min/max:", df["setpoint_clg"].min(), df["setpoint_clg"].max())
    print("comfort_violation min/mean/max:", df["comfort_violation_degCh"].min(),
          df["comfort_violation_degCh"].mean(), df["comfort_violation_degCh"].max())

    print_monthly_tables_extra(df, location)
    print_monthly_tables_split(df, location, time_step_hours=TIME_STEP_HOURS)

    df.to_csv(os.path.join(out_dir, "eval_timeseries.csv"), index=False)

    return df


if __name__ == "__main__":
    bpath, wpath = find_building_and_weather_from_manifest(
        manifest_path,
        location=TARGET_LOCATION,
        occupancy=TARGET_OCCUPANCY,
        thermal=TARGET_THERMAL,
        require_patched=True,
    )

    print("USING BUILDING FILE:", bpath)
    run_eval_for_location(TARGET_LOCATION, str(bpath), str(wpath))