Harmonic Searches

Harmonic searches are searches for ABC, ABCD or XABCD patterns. On the final point of the pattern a price reversal is more likely to occur. The MatrixSearch object performs this search using a very efficent algorithm. A brute force 5 point search would constitute

\[complexity = O(n-m)^5\]

1000 candles with n = 100 peaks would consitiute a 100*99*98*97*96 search with 9034502400 passes!!

MatrixSearch is

\[complexity = O(n^2/2 + (n-m)^2/2)\]

where

\[n\2 <= m <= n\]

1000 candles with n = 100 peaks consitiutes a 100^2/2 ( lower triangle ) scan followed by an additional (n-m)^2/2 loops to link patterns. That’s 5000 + 1250 passes to locate all patterns in the dataframe. That is 1445520 times faster than brute force.

Note

It implies that instead of scanning one Asset in 30 seconds using a brute force approach you can now scan 1.446 Millions assets in the same 30 seconds. The internet is now the slowest part of this problem.

 1>>> from pyharmonics.marketdata import BinanceCandleData
 2>>> from pyharmonics.search import HarmonicSearch
 3>>> from pyharmonics.technicals import Technicals
 4>>> b = BinanceCandleData()
 5>>> b.get_candles('ETHUSDT', b.HOUR_4, 400)
 6>>> t = Technicals(b.df, b.symbol, b.interval)
 7>>> h = HarmonicSearch(t)
 8>>> h.search()
 9>>> patterns = h.get_patterns()
10>>> patterns[m.XABCD]
11[]
12>>> patterns[m.ABCD]
13[ABCDPattern(name='ABCD-50-1.618', formed=True, retraces={'ABC': 0.5000347246336551, 'BCD': 3.31138888888889, 'ABCD': 3.31138888888889}, bullish=False, x=[Timestamp('2023-06-15 12:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-06-17 08:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-06-19 20:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-06-23 20:59:59+0100', tz='Europe/Dublin')], y=[1626.01, 1770.0, 1698.0, 1936.42], abc_extensions=[1936.42], completion_min_price=1930.992, completion_max_price=1930.992)]
14>>> patterns[m.ABCD][0]
15ABCDPattern(name='ABCD-50-1.618', formed=True, retraces={'ABC': 0.5000347246336551, 'BCD': 3.31138888888889, 'ABCD': 3.31138888888889}, bullish=False, x=[Timestamp('2023-06-15 12:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-06-17 08:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-06-19 20:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-06-23 20:59:59+0100', tz='Europe/Dublin')], y=[1626.01, 1770.0, 1698.0, 1936.42], abc_extensions=[1936.42], completion_min_price=1930.992, completion_max_price=1930.992)
16>>> patterns[m.ABC][0]
17ABCPattern(name=0.382, formed=True, retraces={'ABC': 0.386628628131977}, bullish=True, x=[Timestamp('2023-06-15 12:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-07-14 04:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-07-17 20:59:59+0100', tz='Europe/Dublin')], y=[1626.01, 2029.11, 1873.26], abc_extensions=[1873.26], completion_min_price=1873.26, completion_max_price=1873.26)
18>>>

Here we can see a single ABCD pattern formed on ETHUSDT. Its completion time was Timestamp('2023-06-23 20:59:59+0100', tz='Europe/Dublin'). Data can be specifically referenced from the pattern object.

1>>> p = patterns[m.ABC][0]
2>>> p.name
30.382
4>>> p.x
5[Timestamp('2023-06-15 12:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-07-14 04:59:59+0100', tz='Europe/Dublin'), Timestamp('2023-07-17 20:59:59+0100', tz='Europe/Dublin')]
6>>> p.y
7[1626.01, 2029.11, 1873.26]
8>>>

As you can no doubt tell this information can be plotted with b.df to show you where the pattern is on the chart.