Spaces:
Sleeping
Sleeping
File size: 11,641 Bytes
4be2d4d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 | """
์ ์ฅ๋ TensorFlow Lite ๋ชจ๋ธ๊ณผ ์ค์ ์ ์ฌ์ฉํ์ฌ ์ข
๋ชฉ์ ๋ํ ๋ค์ ๋ ์์ธก ์ํ
"""
import os
import sys
import json
import argparse
# NOTE: tensorflow ๋ ๋ฐ๋์ pandas/numpy ๋ณด๋ค ๋จผ์ import ํด์ผ ํ๋ค.
# pandas/numpy ๊ฐ OpenMP ๋ฐํ์(libiomp5md.dll)์ ๋จผ์ ์ฌ๋ฆฌ๋ฉด tensorflow ๋ค์ดํฐ๋ธ
# DLL(_pywrap_tensorflow_internal) ์ด๊ธฐํ๊ฐ ์ค๋ณต ์ถฉ๋ํ์ฌ WinError 1114 ๋ก ์ฃฝ๋๋ค.
import tensorflow as tf
import pandas as pd
import numpy as np
from datetime import datetime, timedelta
from pathlib import Path
import subprocess
import pickle
import warnings
# ๊ฒฝ๊ณ ๋ฌด์
warnings.filterwarnings('ignore', category=UserWarning, module='tensorflow')
# ๋ชจ๋ ๊ฒฝ๋ก ์ถ๊ฐ
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from src.optimization.utils import predict_with_tflite
def get_project_root():
"""ํ๋ก์ ํธ ๋ฃจํธ ๋๋ ํ ๋ฆฌ๋ฅผ ๋ฐํํฉ๋๋ค."""
return Path(__file__).parent.parent
def check_and_data(tickers):
"""
์ฃผ์ ๋ฐ์ดํฐ๊ฐ ์กด์ฌํ๋์ง ํ์ธํ๊ณ , ์์ผ๋ฉด data.py ์คํฌ๋ฆฝํธ๋ก ๊ฐ์ ธ์ต๋๋ค.
"""
ticker_list = tickers.split('_') if '_' in tickers else [tickers]
data_dir = Path(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) / 'data'
data_dir.mkdir(parents=True, exist_ok=True)
missing_tickers = [
ticker for ticker in ticker_list
if not (data_dir / f'{ticker}_data.csv').exists()
]
if not missing_tickers:
return True
print(f"๋๋ฝ๋ ์ข
๋ชฉ ๋ฐ์ดํฐ ๊ฐ์ ธ์ค๊ธฐ: {', '.join(missing_tickers)}")
missing_str = '_'.join(missing_tickers)
try:
script_path = Path(__file__).parent / 'data.py'
result = subprocess.run(
[sys.executable, str(script_path), '--tickers', missing_str],
check=True, capture_output=True, text=True
)
print(result.stdout)
return True
except subprocess.CalledProcessError as e:
print(f"๋ฐ์ดํฐ ๊ฐ์ ธ์ค๊ธฐ ์คํจ: {e}")
print(f"์ค๋ฅ ์ถ๋ ฅ: {e.stderr}")
return False
def load_model_and_config(model_path, config_path):
"""
TensorFlow Lite ๋ชจ๋ธ, ์ค์ , ์ธ์ฝ๋ ์ ๋ณด๋ฅผ ๋ก๋ํฉ๋๋ค.
"""
try:
# ์ค์ ํ์ผ ๋ก๋
with open(config_path, 'r') as f:
config = json.load(f)
# ์ธ์ฝ๋ ์ ๋ณด ๋ก๋
encoders = None
model_path_obj = Path(model_path)
encoder_path = model_path_obj.with_suffix('').with_name(f"{model_path_obj.stem}_encoders.json")
if os.path.exists(encoder_path):
with open(encoder_path, 'r') as f:
encoders = json.load(f)
# TensorFlow Lite ๋ชจ๋ธ ๋ก๋
print(f"๋ชจ๋ธ ๋ก๋ ์ค: {Path(model_path).name}")
interpreter = tf.lite.Interpreter(model_path=str(model_path))
interpreter.allocate_tensors()
return interpreter, config, encoders
except Exception as e:
print(f"๋ชจ๋ธ ๋ก๋ ์คํจ: {e}")
return None, config, encoders
def predict_next_day(model_path, config_path, ticker, output_file='next_day_prediction.csv'):
"""
๋ค์ ๋ ์ฃผ๊ฐ ์์ธก ํจ์
"""
# ๋ชจ๋ธ, ์ค์ , ์ธ์ฝ๋ ๋ก๋
model_or_interpreter, cfg, encoders = load_model_and_config(model_path, config_path)
if model_or_interpreter is None or cfg is None:
print("๋ชจ๋ธ ๋ก๋์ ์คํจํ์ต๋๋ค.")
return None
threshold = cfg.get('best_threshold', cfg.get('threshold', 0.0))
# ํ์ผ ๊ฒฝ๋ก ์ค์
tickers_path = ticker.replace(',', '_') if ',' in ticker else ticker
data_dir = get_project_root() / "data"
processed_dir = data_dir / "processed"
processed_path = processed_dir / f"{tickers_path}_processed.pkl"
# ์ ์ฒ๋ฆฌ๋ ํ์ผ์ด ์์ ๊ฒฝ์ฐ data.py ์คํ
if not processed_path.exists():
try:
script_path = Path(__file__).parent / 'data.py'
result = subprocess.run(
[sys.executable, str(script_path), '--tickers', ticker],
check=True, capture_output=True, text=True
)
except:
return None
# ์ ์ฒ๋ฆฌ๋ ๋ฐ์ดํฐ ๋ก๋
try:
print(f"{ticker} ๋ฐ์ดํฐ ๋ก๋ ์ค...")
with open(processed_path, 'rb') as f:
processed_data = pickle.load(f)
# ๋ฐ์ดํฐ ๋์
๋๋ฆฌ ์ถ์ถ
data_dict = processed_data[0] if isinstance(processed_data, tuple) else processed_data
# ์ํ์ค ๋ฐ์ดํฐ ์ ํ (ํ
์คํธ > ๊ฒ์ฆ > ํ๋ จ ์)
if 'x_test' in data_dict and len(data_dict['x_test']) > 0:
source = 'test'
elif 'x_val' in data_dict and len(data_dict['x_val']) > 0:
source = 'val'
elif 'x_train' in data_dict and len(data_dict['x_train']) > 0:
source = 'train'
else:
return None
# ํ์ํ ๋ฐ์ดํฐ ์ถ์ถ
last_sequence = data_dict[f'x_{source}'][-1:]
last_ticker = data_dict[f'ticker_{source}'][-1]
last_time_diff = data_dict[f'time_diffs_{source}'][-1:]
# ์นํฐ/์ฐ์
์ ๋ณด (์์ผ๋ฉด ์ฌ์ฉ, ์์ผ๋ฉด 0์ผ๋ก ์ค์ )
sector_id = 0
industry_id = 0
if f'sector_{source}' in data_dict:
sector_id = data_dict[f'sector_{source}'][-1]
if f'industry_{source}' in data_dict:
industry_id = data_dict[f'industry_{source}'][-1]
print(f"์์ธก ์ํ ์ค...")
if hasattr(model_or_interpreter, 'predict'):
inputs = [
tf.cast(last_sequence, tf.float32),
tf.cast(np.array([last_ticker]), tf.int32),
tf.cast(np.array([sector_id]), tf.int32),
tf.cast(np.array([industry_id]), tf.int32),
tf.cast(last_time_diff, tf.float32),
]
y_pred_all = model_or_interpreter.predict(inputs, verbose=0)
y_pred = y_pred_all[0] if isinstance(y_pred_all, list) else y_pred_all
else: # TensorFlow Lite ์ธํฐํ๋ฆฌํฐ
# ์
๋ ฅ ํ
์ ์ ๋ณด ๊ฐ์ ธ์ค๊ธฐ
input_details = model_or_interpreter.get_input_details()
# ํ์ํ ๊ฒฝ์ฐ ์
๋ ฅ ๋ฐ์ดํฐ ์ฌ๊ตฌ์ฑ
inputs = []
for i, detail in enumerate(input_details):
name = detail['name'].lower() if hasattr(detail['name'], 'lower') else ""
if 'time' in name:
inputs.append(last_time_diff.astype(np.float32))
elif 'ticker' in name:
inputs.append(np.array([last_ticker], dtype=np.int32))
elif 'industry' in name:
inputs.append(np.array([industry_id], dtype=np.int32))
elif 'sector' in name:
inputs.append(np.array([sector_id], dtype=np.int32))
else:
seq_data = last_sequence.astype(np.float32)
if len(detail['shape']) == 4 and len(seq_data.shape) == 3:
seq_data = np.expand_dims(seq_data, axis=-1)
inputs.append(seq_data)
if not inputs or len(inputs) != len(input_details):
inputs = [
last_time_diff.astype(np.float32),
last_sequence.astype(np.float32),
np.array([last_ticker], dtype=np.int32),
np.array([industry_id], dtype=np.int32),
np.array([sector_id], dtype=np.int32)
]
# TensorFlow Lite ๋ชจ๋ธ๋ก ์์ธก
y_pred = predict_with_tflite(model_or_interpreter, inputs, verbose=False)
if y_pred is None:
return None
# ์์ธก๊ฐ ์ถ์ถ
if isinstance(y_pred, list):
value_output = y_pred[0]
if len(value_output.shape) == 3:
pred_value = float(value_output[0, -1, 0])
elif len(value_output.shape) == 2:
pred_value = float(value_output[0, 0])
else:
pred_value = float(value_output.flatten()[-1])
elif hasattr(y_pred, 'shape'):
if len(y_pred.shape) == 3:
pred_value = float(y_pred[0, -1, 0])
elif y_pred.shape == (1,):
pred_value = float(y_pred[0])
elif y_pred.shape == (1, 1):
pred_value = float(y_pred[0, 0])
else:
pred_value = float(y_pred.flatten()[-1])
else:
pred_value = float(y_pred)
signal = 'BUY' if pred_value > threshold else 'SELL' if pred_value < -threshold else 'HOLD'
confidence = abs(pred_value)
# ๊ฒฐ๊ณผ ์ถ๋ ฅ
print(f"\n===== {ticker} ๋ค์ ๋ ์์ธก =====")
print(f"์์ธก๊ฐ: {pred_value:.6f}")
print(f"์๊ณ๊ฐ: {threshold:.6f}")
print(f"์ ํธ: {signal}")
print(f"์ ๋ขฐ๋: {confidence:.6f}")
# ๊ฒฐ๊ณผ๋ฅผ CSV๋ก ์ ์ฅ
results_df = pd.DataFrame({
'ticker': [ticker],
'prediction_date': [datetime.now().strftime('%Y-%m-%d')],
'predicted_value': [pred_value],
'threshold': [threshold],
'signal': [signal],
'confidence': [confidence]
})
# ๋ชจ๋ธ ๋๋ ํ ๋ฆฌ์ ์ ์ฅ
models_dir = get_project_root() / "models"
models_dir.mkdir(exist_ok=True)
output_path = models_dir / output_file
results_df.to_csv(output_path, index=False)
print(f"๊ฒฐ๊ณผ ์ ์ฅ: {output_path}")
return results_df
except Exception as e:
return None
def main():
parser = argparse.ArgumentParser(description="์ ์ฅ๋ TensorFlow Lite ๋ชจ๋ธ๋ก ๋ค์ ๋ ์ฃผ๊ฐ ์์ธก")
parser.add_argument('--model', type=str, default='models/best_contime_grid_search.tflite',
help='์ ์ฅ๋ TensorFlow Lite ๋ชจ๋ธ ๊ฒฝ๋ก')
parser.add_argument('--config', type=str, default='models/results/best_contime_grid_search_meta.json',
help='์ ์ฅ๋ ์ค์ ํ์ผ ๊ฒฝ๋ก')
parser.add_argument('--tickers', type=str, required=True,
help='์์ธกํ ์ข
๋ชฉ (๋จ์ผ ์ข
๋ชฉ)')
parser.add_argument('--output', type=str, default='predictions.csv',
help='์์ธก ๊ฒฐ๊ณผ ์ ์ฅ ๊ฒฝ๋ก')
args = parser.parse_args()
# ๋ฐ์ดํฐ ํ์ผ์ด ์๋์ง ํ์ธํ๊ณ , ์์ผ๋ฉด ๋ฐ์ดํฐ ๊ฐ์ ธ์ค๊ธฐ
if not check_and_data(args.tickers):
print("๋ฐ์ดํฐ ์ค๋น์ ์คํจํ์ต๋๋ค.")
return
# ๋ค์ค ์ข
๋ชฉ ์ฒ๋ฆฌ ๋ฐฉ์ง
if '_' in args.tickers:
print("๋จ์ผ ์ข
๋ชฉ๋ง ์์ธก ๊ฐ๋ฅํฉ๋๋ค. ์ฌ๋ฌ ์ข
๋ชฉ์ ๊ฐ๋ณ์ ์ผ๋ก ์คํํด์ฃผ์ธ์.")
return
# ๋ค์๋ ์์ธก ์ํ
predict_next_day(
model_path=args.model,
config_path=args.config,
ticker=args.tickers,
output_file=args.output
)
if __name__ == "__main__":
main()
# ----- ํฌํธํด๋ฆฌ์ค ์ฑ๋ฅ -----
# ํ
์คํธ ์ธํธ ์ด ์์ต๋ฅ : 0.2360
# ํ
์คํธ ์ธํธ ์คํ ๋น์จ: 0.0472
# ํ
์คํธ ์ธํธ ์ต๋ ๋ํญ: -0.1322
# ํ
์คํธ ์ธํธ ๊ฑฐ๋ ์: 54
# ----- ๊ฐ๋ณ ์ข
๋ชฉ ํ๊ท ์ฑ๋ฅ -----
# ํ
์คํธ ์ธํธ ํ๊ท ์ข
๋ชฉ ์์ต๋ฅ : 0.2360
# ํ
์คํธ ์ธํธ ํ๊ท ์ข
๋ชฉ ์คํ ๋น์จ: 0.0457 |