Zigzag indicator implementation using numba will close issue #443 and complete pr #693

[aligheshlaghi97]

Hello @twopirllc

I'm pleased to present a new pull request for the completion of the zigzag indicator.
In this update, I've utilized Numba for implementation, as opposed to the previous approach which relied on Pandas. and compared the results with TradingView's default zigzag indicator and everything seems to be so much similar.

Below, you'll find the outcome of applying a deviation of 5% and a pivot leg of 10 to the BTCUSD daily chart:

Compare this with the result of the new implementation under the same configuration:

In order to test it yourself, use the following code:

import sys
import importlib
import numpy as np
from numba import njit
import yfinance as yf
import plotly.graph_objects as go

sys.path.insert(1, '/path-to/pandas-ta')
from pandas_ta import zigzag

ticker = "BTC-USD"
start = "2020-01-01"
interval = "1d"

# Retrieve the stock data using yfinance
data = yf.download(ticker, start, interval=interval)

# Extract the OHLC data
ohlc_data = data.loc[:, ["Open", "High", "Low", "Close"]]

highs = ohlc_data['High']
lows = ohlc_data['Low']

df = zigzag(high=highs, low=lows, pivot_leg=10, price_deviation=5)

high_pivots = df[(df['ZIGZAGt_5%_10'].notna()) & (df['ZIGZAGt_5%_10']==1)]
low_pivots = df[(df['ZIGZAGt_5%_10'].notna()) & (df['ZIGZAGt_5%_10']==-1)]
all_pivots = df[df['ZIGZAGt_5%_10'].notna()]

fig = go.Figure()

fig.add_trace(
    go.Candlestick(
        x=data.index,
        open=ohlc_data["Open"],
        high=ohlc_data["High"],
        low=ohlc_data["Low"],
        close=ohlc_data["Close"],
        name="OHLC",
    )
)

fig.add_trace(go.Scatter(
    x=high_pivots.index, 
    y=high_pivots['ZIGZAGv_5%_10'], 
    mode="markers", 
    name="zigzag highs", 
    marker_color='rgba(0, 152, 0, .8)',
    marker_size=10,
))

fig.add_trace(go.Scatter(
    x=low_pivots.index, 
    y=low_pivots['ZIGZAGv_5%_10'], 
    mode="markers", 
    name="zigzag lows", 
    marker_color='rgba(152, 0, 0, .8)', 
    marker_size=10
))

fig.add_trace(go.Scatter(
    x=all_pivots.index, 
    y=all_pivots['ZIGZAGv_5%_10'], 
    mode="lines", 
    name="zigzag",
    line_color='rgba(0, 0, 152, 0.8)'
))

# Set the layout
fig.update_layout(
    title="ZigZag Indicator Implementations",
    xaxis_rangeslider_visible=False,
    xaxis_title="Date",
    yaxis_title="Price",
    legend=dict(x=0.9, y=1, bgcolor="rgba(255, 255, 255, 0.5)", bordercolor="rgba(0, 0, 0, 0.5)"),
    showlegend=True,
)

# Show the plot
fig.show()

Additionally, I believe this code snippet could prove beneficial for @AureliusMarcusHu and @KilianB in their Pandas-TA coding endeavors.

[twopirllc]

@aligheshlaghi97

Great! Will check it out when I can. 😎

[dmastapkovich]

@aligheshlaghi97

Check the tests, please. That's a very cool improvement.

[aligheshlaghi97]

@dmastapkovich

I reviewed the logs and it seems that the issues aren't stemming from my implementation. Can you confirm, @twopirllc ?

[twopirllc]

@aligheshlaghi97 & @dmastapkovich

Failed Tests

These fails occur after adding/removing indicators, so yes adding zigzag creates the following
errors because the column counts were not updated in test_studies.py. But these are minor issues and a simple fix and I can take care of them after I have reviewed and made any touch ups before merging zigzag into development.

FAILED tests/test_studies.py::test_study_category_columns[trend-30] - assert ...
FAILED tests/test_studies.py::test_study_category_columns[all-323] - assert 3...
FAILED tests/test_studies.py::test_study_all_multirun_talib[False] - assert 6...
FAILED tests/test_studies.py::test_study_all_multirun_talib[True] - assert 62...

I suppose some additional testing code can be added to determine column counts automatically to avoid
these minor fails but it is not a priority.

*Warnings

Lastly, you can ignore the Warnings as most of them will be taken care of on the next development update. The only remaining Warning I have to figure out is:

../env11/lib/python3.11/site-packages/pandas_datareader/compat/__init__.py:11: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
    PANDAS_VERSION = LooseVersion(pd.__version__)

Supposedly it is still an open issue for pandas-datareader until a new release comes out.

[KilianB]

Thank you very much for the improved version. It looks promising and I'll play a around with it a bit more once I have a bit more time.
One small odity in your example above are two low pivots back to back without a high in between

2021-09-07 00:00:00 | -1.0 | 43285.20703125 | 14.256289283356438
2021-09-13 00:00:00 | -1.0 | 43591.3203125 | 17.524664899408478

[einarjohnson]

this is awesome. when will this be merged into the development branch?