python trend line code example

Example 1: add trendline to plot matplotlib

# plot the data itself
pylab.plot(x,y,'o')

# calc the trendline
z = numpy.polyfit(x, y, 1)
p = numpy.poly1d(z)
pylab.plot(x,p(x),"r--")
# the line equation:
print "y=%.6fx+(%.6f)"%(z[0],z[1])

Example 2: trend line stock python

import trendln
# this will serve as an example for security or index closing prices, or low and high prices
import yfinance as yf # requires yfinance - pip install yfinance
tick = yf.Ticker('^GSPC') # S&P500
hist = tick.history(period="max", rounding=True)
mins, maxs = calc_support_resistance(hist[-1000:].Close)
minimaIdxs, pmin, mintrend, minwindows = calc_support_resistance((hist[-1000:].Low, None)) #support only
mins, maxs = calc_support_resistance((hist[-1000:].Low, hist[-1000:].High))
(minimaIdxs, pmin, mintrend, minwindows), (maximaIdxs, pmax, maxtrend, maxwindows) = mins, maxs

(minimaIdxs, pmin, mintrend, minwindows), (maximaIdxs, pmax, maxtrend, maxwindows) = \
	calc_support_resistance(
	# list/numpy ndarray/pandas Series of data as bool/int/float and if not a list also unsigned
	# or 2-tuple (support, resistance) where support and resistance are 1-dimensional array-like or one or the other is None
	# can calculate only support, only resistance, both for different data, or both for identical data
	h,

	# METHOD_NAIVE - any local minima or maxima only for a single interval (currently requires pandas)
	# METHOD_NAIVECONSEC - any local minima or maxima including those for consecutive constant intervals (currently requires pandas)
	# METHOD_NUMDIFF (default) - numerical differentiation determined local minima or maxima (requires findiff)
	extmethod = METHOD_NUMDIFF,
	
	# METHOD_NCUBED - simple exhuastive 3 point search (slowest)
	# METHOD_NSQUREDLOGN (default) - 2 point sorted slope search (fast)
	# METHOD_HOUGHPOINTS - Hough line transform optimized for points
	# METHOD_HOUGHLINES - image-based Hough line transform (requires scikit-image)
	# METHOD_PROBHOUGH - image-based Probabilistic Hough line transform (requires scikit-image)
	method=METHOD_NSQUREDLOGN,
	
	# window size when searching for trend lines prior to merging together
	window=125,
	
	# maximum percentage slope standard error
	errpct = 0.005,
	
	# for all METHOD_*HOUGH*, the smallest unit increment for discretization e.g. cents/pennies 0.01
	hough_scale=0.01
	
	# only for METHOD_PROBHOUGH, number of iterations to run
	hough_prob_iter=10,
	
	# sort by area under wrong side of curve, otherwise sort by slope standard error
	sortError=False,
	
	# accuracy if using METHOD_NUMDIFF for example 5-point stencil is accuracy=3
	accuracy=1)
# if h is a 2-tuple with one value as None, then a 2-tuple is not returned, but the appropriate tuple instead
# minimaIdxs - sorted list of indexes to the local minima
# pmin - [slope, intercept] of average best fit line through all local minima points
# mintrend - sorted list containing (points, result) for local minima trend lines
	# points - list of indexes to points in trend line
	# result - (slope, intercept, SSR, slopeErr, interceptErr, areaAvg)
		# slope - slope of best fit trend line
		# intercept - y-intercept of best fit trend line
		# SSR - sum of squares due to regression
		# slopeErr - standard error of slope
		# interceptErr - standard error of intercept
		# areaAvg - Reimann sum area of difference between best fit trend line
		#   and actual data points averaged per time unit
# minwindows - list of windows each containing mintrend for that window

# maximaIdxs - sorted list of indexes to the local maxima
# pmax - [slope, intercept] of average best fit line through all local maxima points
# maxtrend - sorted list containing (points, result) for local maxima trend lines
	#see for mintrend above
# maxwindows - list of windows each containing maxtrend for that window