Bug Report: Errors occur for NaN columns.

issue case

When animating over a limited visual range, as in this example, the possibility arises that the signal columns are all NaN.

https://github.com/matplotlib/mplfinance/assets/37892107/a5afea85-57ce-4282-ad7c-955240c54787

issue

mplfinance\plotting.py", line 1108, in _addplot_columns
ymhi = math.log(max(math.fabs(np.nanmax(yd)),1e-7),10)
This Numpy code generates an error.
"ValueError: zero-size array to reduction operation maximum which has no identity"

Error reproduction

import datetime

import numpy as np
import pandas as pd
import mplfinance as mpf

# ランダムなOHLCデータを生成する関数
def generate_sample_ohlc_data(n=10):
    # 現在の日付を取得
    base_date = datetime.datetime.now()
    # 日付のリストを生成
    dates = [base_date - datetime.timedelta(days=i) for i in range(n)]
    dates.reverse()

    # ランダムな価格データを生成
    rand = np.random.default_rng().uniform
    open = [100]
    high = [100.02]
    low = [100.02]
    close = [100.05]
    for i in range(n-1):
        open.append(close[i])
        close.append(open[-1] + rand(-0.1, 0.1))
        high.append(max(open[-1],close[-1]) + rand(0, 0.05))
        low.append(min(open[-1],close[-1]) - rand(0, 0.05))

    # DataFrameに格納
    ohlc_data = pd.DataFrame({
        "date": dates,
        "open": open,
        "high": high,
        "low": low,
        "close": close
    })

    return ohlc_data.set_index("date", drop=True)

period_range = 50
df = generate_sample_ohlc_data(period_range)
df["signal"] = np.nan
print(df)

some_signal = mpf.make_addplot(df["signal"])
mpf.plot(df, type='candle', addplot=some_signal) # ValueError

Currently, we have worked around this by doing NaN pre-checking on the user side, but it would be great if the library could handle this.

Comienza con la reproducción del problema en el issue e inspecciona mplfinance/plotting.py en _addplot_columns, especialmente la llamada a np.nanmax alrededor de la línea 1108. Confirma el fallo con una columna de addplot que contenga únicamente valores NaN; se considera terminado cuando graficar esos datos ya no provoca el ValueError de reducción de tamaño cero.