model2 / app.py
kaganseyda's picture
Update app.py
91c96e5 verified
"""
Robust TALib + mplfinance + Gradio example
- Normalizes column names (handles tuple/MultiIndex columns)
- Keeps Volume if present
- Converts everything numeric and drops NaN OHLC rows
- Calls TALib pattern functions safely and plots with mplfinance
"""
import yfinance as yf
import pandas as pd
import numpy as np
import mplfinance as mpf
import talib
import gradio as gr
from datetime import date
import os
from typing import Optional, Tuple, List
# Pattern list from talib (CDL* functions)
TALIB_PATTERNS = sorted([name for name in dir(talib) if name.startswith("CDL")])
def _normalize_col_name(col) -> str:
"""Turn any column key into a simple lowercase string.
Handles tuples (MultiIndex) by joining parts with '_'."""
if isinstance(col, (tuple, list)):
# join non-empty parts, convert to string
parts = [str(c) for c in col if c is not None]
joined = "_".join(parts)
return joined.strip().lower()
return str(col).strip().lower()
def _find_best_col(key: str, columns: List[str]) -> Optional[str]:
"""Given normalized columns, find best matching column for key (open/high/low/close/volume)."""
key = key.lower()
# Exact match
if key in columns:
return key
# Prefer suffix matches (e.g., msft_open)
for c in columns:
if c.endswith("_" + key):
return c
# Fallback: any column containing the key
for c in columns:
if key in c:
return c
return None
def clean_ohlc(df: pd.DataFrame) -> pd.DataFrame:
"""
Normalize column names, find OHLC (and optional Volume), convert to numeric,
drop NaNs and ensure DatetimeIndex. Returns a DataFrame with columns:
['Open','High','Low','Close'] and optionally 'Volume'.
Raises ValueError on irrecoverable problems.
"""
if df is None:
raise ValueError("Gelen dataframe None.")
if not isinstance(df, pd.DataFrame):
raise ValueError(f"Gelen obje DataFrame değil: {type(df)}")
# Make a copy to avoid mutating caller's df
df_work = df.copy()
# Flatten column names into simple lowercase strings
try:
normalized_columns = [_normalize_col_name(c) for c in df_work.columns]
except Exception as e:
# worst-case fallback: convert all to string then normalize
normalized_columns = [str(c).strip().lower() for c in df_work.columns]
# assign normalized columns temporarily (we'll keep mapping)
col_map = dict(zip(normalized_columns, df_work.columns))
df_work.columns = normalized_columns
# find required columns
required_keys = ['open', 'high', 'low', 'close']
found = {}
for k in required_keys:
matched = _find_best_col(k, normalized_columns)
if matched is None:
raise ValueError(f"Veride '{k}' sütunu bulunamadı. Mevcut sütunlar: {list(normalized_columns)}")
found[k] = matched
# optional volume
vol_key = _find_best_col('volume', normalized_columns)
include_volume = vol_key is not None
# select and rename to canonical names
cols_to_take = [found['open'], found['high'], found['low'], found['close']]
if include_volume:
cols_to_take.append(vol_key)
df_sel = df_work[cols_to_take].copy()
rename_map = {
found['open']: 'Open',
found['high']: 'High',
found['low']: 'Low',
found['close']: 'Close'
}
if include_volume:
rename_map[vol_key] = 'Volume'
df_sel = df_sel.rename(columns=rename_map)
# index -> DatetimeIndex if possible
if not isinstance(df_sel.index, pd.DatetimeIndex):
# try 'date' column if exists
if 'date' in df_sel.columns:
try:
df_sel.index = pd.to_datetime(df_sel['date'], errors='coerce')
df_sel = df_sel.drop(columns=['date'])
except Exception:
pass
else:
# try to parse existing index
try:
df_sel.index = pd.to_datetime(df_sel.index, errors='coerce')
except Exception:
pass
# convert to numeric
numeric_cols = ['Open', 'High', 'Low', 'Close']
if include_volume:
numeric_cols.append('Volume')
df_sel[numeric_cols] = df_sel[numeric_cols].apply(pd.to_numeric, errors='coerce')
# drop rows with NaN OHLC or NaT index
if not isinstance(df_sel.index, pd.DatetimeIndex):
# if index still not datetime, try to reset index and parse date column
df_sel = df_sel.reset_index()
if 'index' in df_sel.columns:
try:
df_sel['index'] = pd.to_datetime(df_sel['index'], errors='coerce')
df_sel = df_sel.set_index('index')
except Exception:
pass
# drop any rows with NaT index now
if isinstance(df_sel.index, pd.DatetimeIndex):
df_sel = df_sel[~df_sel.index.isna()]
df_sel = df_sel.dropna(subset=['Open', 'High', 'Low', 'Close']).copy()
if df_sel.empty:
raise ValueError("value error.")
# ensure ordering ascending by date
df_sel = df_sel.sort_index()
return df_sel
def find_candlestick_patterns(df: pd.DataFrame, pattern_name: str) -> Tuple[List, Optional[str]]:
"""
Run TALib pattern and return list of addplots (apds) for mplfinance,
or ([], error_message) on error.
"""
# check callable pattern
try:
pattern_func = getattr(talib, pattern_name)
except AttributeError:
return [], f"Error: '{pattern_name}' formasyonu TALib'te bulunamadı."
if not callable(pattern_func):
return [], f"Error: '{pattern_name}' TALib içinde callable değil."
# clean and prepare data
try:
df_clean = clean_ohlc(df)
except Exception as e:
return [], f"Veri temizleme hatası: {e}"
try:
open_arr = np.asarray(df_clean['Open'].values, dtype=float).ravel()
high_arr = np.asarray(df_clean['High'].values, dtype=float).ravel()
low_arr = np.asarray(df_clean['Low'].values, dtype=float).ravel()
close_arr= np.asarray(df_clean['Close'].values, dtype=float).ravel()
# sanity check lengths
n = len(open_arr)
if not (len(high_arr) == len(low_arr) == len(close_arr) == n):
return [], "Error: OHLC array uzunlukları eşit değil."
except Exception as e:
return [], f"Array dönüşüm hatası: {e}"
# call talib
try:
pattern_result = pattern_func(open_arr, high_arr, low_arr, close_arr)
pattern_series = pd.Series(pattern_result, index=df_clean.index)
except Exception as e:
return [], f"TALib çalıştırma hatası: {e}"
apds = []
# bullish (positive)
bull_idx = pattern_series[pattern_series > 0].index
if len(bull_idx) > 0:
bullish_points = pd.Series(np.nan, index=df_clean.index)
# vectorized assignment for indices
bullish_points.loc[bull_idx] = df_clean.loc[bull_idx, 'Low'] * 0.98
apds.append(
mpf.make_addplot(
bullish_points,
type='scatter',
marker='^',
markersize=60,
color='green',
panel=0,
alpha=0.85
)
)
# bearish (negative)
bear_idx = pattern_series[pattern_series < 0].index
if len(bear_idx) > 0:
bearish_points = pd.Series(np.nan, index=df_clean.index)
bearish_points.loc[bear_idx] = df_clean.loc[bear_idx, 'High'] * 1.02
apds.append(
mpf.make_addplot(
bearish_points,
type='scatter',
marker='v',
markersize=60,
color='red',
panel=0,
alpha=0.85
)
)
return apds, None
def plot_stock_with_patterns(ticker_symbol: str, start_date: str, end_date: str, selected_patterns) -> Tuple[Optional[str], str]:
"""
Main handler for Gradio. Returns (image_filepath or None, status_message).
"""
if not ticker_symbol:
return None, "Error: stock symbol could not empty."
try:
start = pd.to_datetime(start_date)
end = pd.to_datetime(end_date)
if start >= end:
return None, "Error: date error."
except Exception:
return None, "Error: invalid date format"
# download
try:
df = yf.download(ticker_symbol, start=start_date, end=end_date, progress=False)
if df is None or (isinstance(df, pd.DataFrame) and df.empty):
return None, f"Error: '{ticker_symbol}' için veri bulunamadı (yfinance boş döndü)."
except Exception as e:
return None, f"Veri indirilirken hata oluştu: {e}"
# Clean once here to feed to mplfinance and TALib
try:
df_clean = clean_ohlc(df)
except Exception as e:
return None, f"Veri temizleme hatası: {e}"
# patterns
all_apds = []
if selected_patterns:
# selected_patterns might be tuple/list/str
if isinstance(selected_patterns, str):
selected_patterns = [selected_patterns]
for pattern_name in selected_patterns:
pattern_apds, err = find_candlestick_patterns(df_clean, pattern_name)
if err:
return None, err
all_apds.extend(pattern_apds)
# prepare fig path
os.makedirs('/tmp', exist_ok=True)
safe_ticker = str(ticker_symbol).replace("/", "_").replace("\\", "_")
fig_path = f"/tmp/stock_chart_{safe_ticker}_{pd.Timestamp.now().strftime('%Y%m%d%H%M%S')}.png"
# style
s = mpf.make_mpf_style(base_mpf_style='yahoo', mavcolors=['#1f77b4', '#ff7f0e', '#2ca02c'])
# detect if volume exists
has_volume = 'Volume' in df_clean.columns
try:
fig, _ = mpf.plot(
df_clean,
type='candle',
volume=has_volume,
addplot=all_apds if all_apds else None,
style=s,
title=f"\n{ticker_symbol} Candlestick Chart",
ylabel='Price',
ylabel_lower='Volume' if has_volume else None,
returnfig=True,
figscale=1.5
)
fig.savefig(fig_path, dpi=150, bbox_inches='tight')
return fig_path, "Successfully created!"
except Exception as e:
return None, f"error: {e}"
# Gradio arayüzü
iface = gr.Interface(
fn=plot_stock_with_patterns,
inputs=[
gr.Textbox(label="Stock Symbol (örn: MSFT, AAPL, ^GSPC)", value="MSFT"),
gr.Textbox(label="Starting Date (YYYY-MM-DD)", value="2024-01-01"),
gr.Textbox(label="End Date (YYYY-MM-DD)", value=date.today().strftime("%Y-%m-%d")),
gr.CheckboxGroup(
label="Select one (TALib)",
choices=TALIB_PATTERNS,
value=["CDLHAMMER", "CDLSHOOTINGSTAR"]
)
],
outputs=[
gr.Image(type="filepath", label="Hisse Grafiği"),
gr.Textbox(label="Durum")
],
title="TALib Tool",
description="bullish green, bearish red"
)
if __name__ == "__main__":
iface.launch()