kimyechan
commited on
Commit
·
7355d65
1
Parent(s):
0b16dd0
fix:수정
Browse files
app.py
CHANGED
|
@@ -3,6 +3,7 @@ import pandas as pd
|
|
| 3 |
import torch
|
| 4 |
import gradio as gr
|
| 5 |
import yfinance as yf
|
|
|
|
| 6 |
|
| 7 |
from chronos import BaseChronosPipeline # pip: chronos-forecasting
|
| 8 |
|
|
@@ -24,17 +25,98 @@ def get_pipeline(model_id: str, device: str = "cpu"):
|
|
| 24 |
|
| 25 |
|
| 26 |
# =============================
|
| 27 |
-
#
|
| 28 |
# =============================
|
| 29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
"""
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
"""
|
| 39 |
ticker = ticker.strip().upper()
|
| 40 |
|
|
@@ -45,12 +127,12 @@ def load_close_series(ticker: str, start: str, end: str, interval: str = "1d"):
|
|
| 45 |
sdt = pd.to_datetime(_start)
|
| 46 |
edt = pd.to_datetime(_end)
|
| 47 |
if edt < sdt:
|
| 48 |
-
sdt, edt = edt, sdt
|
| 49 |
_start, _end = sdt.date().isoformat(), edt.date().isoformat()
|
| 50 |
except Exception:
|
| 51 |
-
pass
|
| 52 |
|
| 53 |
-
def _extract_close(df):
|
| 54 |
if df is None or df.empty:
|
| 55 |
return None
|
| 56 |
c = df.get("Close")
|
|
@@ -85,12 +167,11 @@ def load_close_series(ticker: str, start: str, end: str, interval: str = "1d"):
|
|
| 85 |
if interval == "1d":
|
| 86 |
period_candidates = ["max", "10y", "5y", "2y", "1y"]
|
| 87 |
elif interval == "1h":
|
| 88 |
-
period_candidates = ["730d", "365d", "60d"]
|
| 89 |
-
else: # 30m/15m/5m
|
| 90 |
-
period_candidates = ["60d", "30d", "14d"]
|
| 91 |
|
| 92 |
for per in period_candidates:
|
| 93 |
-
# Ticker().history(period=…)
|
| 94 |
try:
|
| 95 |
df = tk.history(period=per, interval=interval, auto_adjust=True, actions=False)
|
| 96 |
s = _extract_close(df)
|
|
@@ -98,7 +179,6 @@ def load_close_series(ticker: str, start: str, end: str, interval: str = "1d"):
|
|
| 98 |
return s
|
| 99 |
except Exception:
|
| 100 |
pass
|
| 101 |
-
# download(period=…)
|
| 102 |
try:
|
| 103 |
df = yf.download(
|
| 104 |
ticker, period=per, interval=interval,
|
|
@@ -110,9 +190,15 @@ def load_close_series(ticker: str, start: str, end: str, interval: str = "1d"):
|
|
| 110 |
except Exception:
|
| 111 |
pass
|
| 112 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
raise ValueError(
|
| 114 |
-
"
|
| 115 |
-
"간격(interval)이나 기간(start/end 혹은 period)을 조정해 다시 시도해 보세요."
|
| 116 |
)
|
| 117 |
|
| 118 |
|
|
@@ -123,26 +209,25 @@ def run_forecast(ticker, start_date, end_date, horizon, model_id, device, interv
|
|
| 123 |
try:
|
| 124 |
series = load_close_series(ticker, start_date, end_date, interval)
|
| 125 |
except Exception as e:
|
| 126 |
-
# Gradio v4: Plot.update 없음 → None 반환
|
| 127 |
return None, pd.DataFrame(), f"데이터 로딩 오류: {e}"
|
| 128 |
|
| 129 |
pipe = get_pipeline(model_id, device)
|
| 130 |
H = int(horizon)
|
| 131 |
|
| 132 |
-
# Chronos
|
| 133 |
context = torch.tensor(series.values, dtype=torch.float32)
|
| 134 |
|
| 135 |
# 예측: (num_series=1, num_quantiles=3, H) with q=[0.1, 0.5, 0.9]
|
| 136 |
preds = pipe.predict(context=context, prediction_length=H)[0]
|
| 137 |
q10, q50, q90 = preds[0], preds[1], preds[2]
|
| 138 |
|
| 139 |
-
# 표
|
| 140 |
df_fcst = pd.DataFrame(
|
| 141 |
{"q10": q10.numpy(), "q50": q50.numpy(), "q90": q90.numpy()},
|
| 142 |
index=pd.RangeIndex(1, H + 1, name="step"),
|
| 143 |
)
|
| 144 |
|
| 145 |
-
# 미래 x
|
| 146 |
import matplotlib.pyplot as plt
|
| 147 |
freq_map = {"1d": "D", "1h": "H", "30m": "30T", "15m": "15T", "5m": "5T"}
|
| 148 |
freq = freq_map.get(interval, "D")
|
|
|
|
| 3 |
import torch
|
| 4 |
import gradio as gr
|
| 5 |
import yfinance as yf
|
| 6 |
+
import requests
|
| 7 |
|
| 8 |
from chronos import BaseChronosPipeline # pip: chronos-forecasting
|
| 9 |
|
|
|
|
| 25 |
|
| 26 |
|
| 27 |
# =============================
|
| 28 |
+
# Binance (무인증) 폴백
|
| 29 |
# =============================
|
| 30 |
+
_BINANCE_INTERVAL = {"1d": "1d", "1h": "1h", "30m": "30m", "15m": "15m", "5m": "5m"}
|
| 31 |
+
|
| 32 |
+
def _yf_to_binance_symbol(ticker: str) -> str | None:
|
| 33 |
+
"""
|
| 34 |
+
BTC-USD -> BTCUSDT, ETH-USD -> ETHUSDT ...
|
| 35 |
+
주식/원화/기타 심볼은 None (Binance 폴백하지 않음)
|
| 36 |
+
"""
|
| 37 |
+
t = ticker.upper()
|
| 38 |
+
if t.endswith("-USD") and len(t) >= 6:
|
| 39 |
+
base = t[:-4] # remove "-USD"
|
| 40 |
+
return f"{base}USDT"
|
| 41 |
+
return None
|
| 42 |
+
|
| 43 |
+
def _fetch_binance_klines(ticker: str, interval: str, start: str | None, end: str | None) -> pd.Series:
|
| 44 |
"""
|
| 45 |
+
Binance Klines (무인증)
|
| 46 |
+
https://api.binance.com/api/v3/klines
|
| 47 |
+
반환: pandas.Series(index=datetime, values=float close)
|
| 48 |
+
"""
|
| 49 |
+
if interval not in _BINANCE_INTERVAL:
|
| 50 |
+
raise ValueError("Binance는 해당 interval을 지원하지 않습니다.")
|
| 51 |
+
|
| 52 |
+
symbol = _yf_to_binance_symbol(ticker)
|
| 53 |
+
if not symbol:
|
| 54 |
+
raise ValueError("이 티커는 Binance 폴백 대상이 아닙니다.")
|
| 55 |
+
|
| 56 |
+
base = "https://api.binance.com/api/v3/klines"
|
| 57 |
+
|
| 58 |
+
def to_ms(s: str) -> int:
|
| 59 |
+
return int(pd.to_datetime(s).timestamp() * 1000)
|
| 60 |
+
|
| 61 |
+
start_ms = to_ms(start) if start else None
|
| 62 |
+
end_ms = to_ms(end) if end else None
|
| 63 |
+
|
| 64 |
+
rows = []
|
| 65 |
+
cur_start = start_ms
|
| 66 |
+
while True:
|
| 67 |
+
params = {"symbol": symbol, "interval": _BINANCE_INTERVAL[interval], "limit": 1000}
|
| 68 |
+
if cur_start is not None:
|
| 69 |
+
params["startTime"] = cur_start
|
| 70 |
+
if end_ms is not None:
|
| 71 |
+
params["endTime"] = end_ms
|
| 72 |
+
|
| 73 |
+
r = requests.get(base, params=params, timeout=30)
|
| 74 |
+
r.raise_for_status()
|
| 75 |
+
data = r.json()
|
| 76 |
+
if not data:
|
| 77 |
+
break
|
| 78 |
+
|
| 79 |
+
rows.extend(data)
|
| 80 |
+
|
| 81 |
+
last_close_time = data[-1][6] # closeTime (ms)
|
| 82 |
+
next_start = last_close_time + 1
|
| 83 |
+
if cur_start is not None and next_start <= cur_start:
|
| 84 |
+
break
|
| 85 |
+
cur_start = next_start
|
| 86 |
+
|
| 87 |
+
if len(data) < 1000:
|
| 88 |
+
break
|
| 89 |
+
|
| 90 |
+
if not rows:
|
| 91 |
+
raise ValueError("Binance에서 데이터가 비어 있습니다.")
|
| 92 |
+
|
| 93 |
+
df = pd.DataFrame(rows, columns=[
|
| 94 |
+
"openTime","open","high","low","close","volume","closeTime",
|
| 95 |
+
"quoteAssetVolume","numTrades","takerBuyBase","takerBuyQuote","ignore"
|
| 96 |
+
])
|
| 97 |
+
df["ts"] = pd.to_datetime(df["closeTime"], unit="ms")
|
| 98 |
+
s = df.set_index("ts")["close"].astype(float).sort_index()
|
| 99 |
+
|
| 100 |
+
if start:
|
| 101 |
+
s = s[s.index >= pd.to_datetime(start)]
|
| 102 |
+
if end:
|
| 103 |
+
s = s[s.index <= pd.to_datetime(end)]
|
| 104 |
+
if s.empty:
|
| 105 |
+
raise ValueError("Binance 시리즈가 비어 있습니다.")
|
| 106 |
+
return s
|
| 107 |
+
|
| 108 |
+
|
| 109 |
+
# =============================
|
| 110 |
+
# yfinance 전용 견고 로더 (+Binance 폴백)
|
| 111 |
+
# =============================
|
| 112 |
+
def load_close_series(ticker: str, start: str, end: str, interval: str = "1d") -> pd.Series:
|
| 113 |
+
"""
|
| 114 |
+
yfinance만으로 먼저 시도:
|
| 115 |
+
1) Ticker().history(start/end)
|
| 116 |
+
2) download(start/end, repair=True)
|
| 117 |
+
3) period 폴백 (interval별 후보 순회)
|
| 118 |
+
그래도 실패할 경우:
|
| 119 |
+
4) Binance (무인증) 폴백 — BTC-USD 같은 암호화폐만 대상
|
| 120 |
"""
|
| 121 |
ticker = ticker.strip().upper()
|
| 122 |
|
|
|
|
| 127 |
sdt = pd.to_datetime(_start)
|
| 128 |
edt = pd.to_datetime(_end)
|
| 129 |
if edt < sdt:
|
| 130 |
+
sdt, edt = edt, sdt
|
| 131 |
_start, _end = sdt.date().isoformat(), edt.date().isoformat()
|
| 132 |
except Exception:
|
| 133 |
+
pass
|
| 134 |
|
| 135 |
+
def _extract_close(df: pd.DataFrame | None) -> pd.Series | None:
|
| 136 |
if df is None or df.empty:
|
| 137 |
return None
|
| 138 |
c = df.get("Close")
|
|
|
|
| 167 |
if interval == "1d":
|
| 168 |
period_candidates = ["max", "10y", "5y", "2y", "1y"]
|
| 169 |
elif interval == "1h":
|
| 170 |
+
period_candidates = ["730d", "365d", "60d"]
|
| 171 |
+
else: # 30m/15m/5m
|
| 172 |
+
period_candidates = ["60d", "30d", "14d"]
|
| 173 |
|
| 174 |
for per in period_candidates:
|
|
|
|
| 175 |
try:
|
| 176 |
df = tk.history(period=per, interval=interval, auto_adjust=True, actions=False)
|
| 177 |
s = _extract_close(df)
|
|
|
|
| 179 |
return s
|
| 180 |
except Exception:
|
| 181 |
pass
|
|
|
|
| 182 |
try:
|
| 183 |
df = yf.download(
|
| 184 |
ticker, period=per, interval=interval,
|
|
|
|
| 190 |
except Exception:
|
| 191 |
pass
|
| 192 |
|
| 193 |
+
# 4) Binance 폴백 (암호화폐만)
|
| 194 |
+
try:
|
| 195 |
+
s = _fetch_binance_klines(ticker, interval, _start, _end)
|
| 196 |
+
return s
|
| 197 |
+
except Exception:
|
| 198 |
+
pass
|
| 199 |
+
|
| 200 |
raise ValueError(
|
| 201 |
+
"데이터를 가져오지 못했습니다. 간격(interval)이나 기간(start/end 혹은 period)을 조정해 다시 시도해 보세요."
|
|
|
|
| 202 |
)
|
| 203 |
|
| 204 |
|
|
|
|
| 209 |
try:
|
| 210 |
series = load_close_series(ticker, start_date, end_date, interval)
|
| 211 |
except Exception as e:
|
|
|
|
| 212 |
return None, pd.DataFrame(), f"데이터 로딩 오류: {e}"
|
| 213 |
|
| 214 |
pipe = get_pipeline(model_id, device)
|
| 215 |
H = int(horizon)
|
| 216 |
|
| 217 |
+
# Chronos 입력
|
| 218 |
context = torch.tensor(series.values, dtype=torch.float32)
|
| 219 |
|
| 220 |
# 예측: (num_series=1, num_quantiles=3, H) with q=[0.1, 0.5, 0.9]
|
| 221 |
preds = pipe.predict(context=context, prediction_length=H)[0]
|
| 222 |
q10, q50, q90 = preds[0], preds[1], preds[2]
|
| 223 |
|
| 224 |
+
# 표
|
| 225 |
df_fcst = pd.DataFrame(
|
| 226 |
{"q10": q10.numpy(), "q50": q50.numpy(), "q90": q90.numpy()},
|
| 227 |
index=pd.RangeIndex(1, H + 1, name="step"),
|
| 228 |
)
|
| 229 |
|
| 230 |
+
# 미래 x축
|
| 231 |
import matplotlib.pyplot as plt
|
| 232 |
freq_map = {"1d": "D", "1h": "H", "30m": "30T", "15m": "15T", "5m": "5T"}
|
| 233 |
freq = freq_map.get(interval, "D")
|