Trendline Regression TV -> PRT

This topic has 3 replies, 3 voices, and was last updated 11 months ago by LucasBest.

Viewing 4 posts - 1 through 4 (of 4 total)

10/30/2023 at 11:21 AM #223046

ashehzi

Hi PRC team, hope you guys are well. Thanks for whatever you have been doing wrt PRT/PRC.

I’ve used Trend Line Regression indicator on TradingView (https://www.tradingview.com/script/tf2TTN6c/) and I think we PRT community will surely benefit from it. I appreciate is someone could convert it. As the source code is open, I am sharing here. It is combo of a shared library and the indicator code using that library.

// This source code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/
// © dandrideng

//@version=5

// @description least_squares_regression: Least squares regression algorithm
library("least_squares_regression", overlay=true)

// @function basic_lsr: Basic least squares regression algorithm
// @param series int[] t: time scale value array corresponding to price
// @param series float[] p: price scale value array corresponding to time
// @param series int array_size: the length of regression array
// @returns reg_slop, reg_intercept, reg_level, reg_stdev
export basic_lsr(series int[] t, series float[] p, series int array_size) =>
	p_t = array.new_float(array_size, 0)
	t_2 = array.new_float(array_size, 0)
	
	//init array
	for i = 0 to array_size - 1
	    array.set(p_t, i, array.get(p, i) * array.get(t, i))
	    array.set(t_2, i, array.get(t, i) * array.get(t, i))
	
	//get slop and intercept
	sum_t = array.sum(t)
	sum_p = array.sum(p)
	sum_t_2 = array.sum(t_2)
	sum_p_t = array.sum(p_t)
	a = (array_size * sum_p_t - sum_p * sum_t) / (array_size * sum_t_2 - math.pow(sum_t, 2))
	b = (sum_p - a * sum_t) / array_size
    
    //get regression prediction value
    _p = array.new_float(array_size, 0)
    for i = 0 to array_size - 1
        array.set(_p, i, a * array.get(t, i) + b)
    
    //get regression level 
    float r1 = 0
    float r2 = 0
    p_avg = array.avg(p)
    for i = 0 to array_size - 1
        r1 += math.pow(array.get(_p, i) - p_avg, 2)
        r2 += math.pow(array.get(p, i) - p_avg, 2)
    r = -math.log10(1 - r1 / r2)
    
    //get regression standard devition
    _s = array.new_float(array_size, 0)
    for i = 0 to array_size - 1
        array.set(_s, i, array.get(p, i) - array.get(_p, i))
    s = array.stdev(_s)
	
	//return results
	[a, b, r, s]

// @function top_trend_line_lsr: Trend line fitting based on least square algorithm
// @param series int[] t: time scale value array corresponding to price
// @param series float[] p: price scale value array corresponding to time
// @param series int array_size: the length of regression array
// @param string reg_type: regression type in 'top' and 'bottom'
// @param series int max_iter: maximum fitting iterations
// @param series int min_points: the threshold of regression point numbers
// @returns reg_slop, reg_intercept, reg_level, reg_stdev, reg_point_num
export trend_line_lsr(series int[] t, series float[] p, series int array_size, string reg_type, series int max_iter, series int min_points) =>
	[a, b, r, s] = basic_lsr(t, p, array_size)
	x1 = array.get(t, 0)
	x2 = array.get(t, array_size - 1)
	n = array_size //Function arguments cannot be mutable, what the fuck!

	for i = 0 to max_iter - 1
	    //exit conditions
    	if n < min_points
    	    break
    	   
    	//init new tick and price array
        _t = array.new_int(n, 0)
        _p = array.new_float(n, 0)
        _n = 0
        
        //fine the ground truth values that bigger than predictions
        for j = 0 to n - 1
            if reg_type == 'top'
                if (a * array.get(t, j) + b) < array.get(p, j)
                    array.set(_t, _n, array.get(t, j))
                    array.set(_p, _n, array.get(p, j))
                    _n += 1
            else if reg_type == 'bottom'
                if (a * array.get(t, j) + b) > array.get(p, j)
                    array.set(_t, _n, array.get(t, j))
                    array.set(_p, _n, array.get(p, j))
                    _n += 1
            else
                break
        
        //exit if new array size is less than threshold
        if _n < min_points
            break
        
        //override result if r of new array is bigger than last array
        [_a, _b, _r, _s] = basic_lsr(_t, _p, _n)
        if _r > r
            x1 := array.get(_t, 0)
            x2 := array.get(_t, _n - 1)
            a := _a
            b := _b
            r := _r
            s := _s
            n := _n
    
    //after for loop 
    [x1, x2, a, b, r, s, n]

100

101

102

103

104

105

106

107

108

109

// This source code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/

// © dandrideng

//@version=5

// @description least_squares_regression: Least squares regression algorithm

library("least_squares_regression", overlay=true)

// @function basic_lsr: Basic least squares regression algorithm

// @param series int[] t: time scale value array corresponding to price

// @param series float[] p: price scale value array corresponding to time

// @param series int array_size: the length of regression array

// @returns reg_slop, reg_intercept, reg_level, reg_stdev

export basic_lsr(series int[] t, series float[] p, series int array_size) =>

p_t = array.new_float(array_size, 0)

t_2 = array.new_float(array_size, 0)

//init array

for i = 0 to array_size - 1

array.set(p_t, i, array.get(p, i) * array.get(t, i))

array.set(t_2, i, array.get(t, i) * array.get(t, i))

//get slop and intercept

sum_t = array.sum(t)

sum_p = array.sum(p)

sum_t_2 = array.sum(t_2)

sum_p_t = array.sum(p_t)

a = (array_size * sum_p_t - sum_p * sum_t) / (array_size * sum_t_2 - math.pow(sum_t, 2))

b = (sum_p - a * sum_t) / array_size

//get regression prediction value

_p = array.new_float(array_size, 0)

for i = 0 to array_size - 1

array.set(_p, i, a * array.get(t, i) + b)

//get regression level

float r1 = 0

float r2 = 0

p_avg = array.avg(p)

for i = 0 to array_size - 1

r1 += math.pow(array.get(_p, i) - p_avg, 2)

r2 += math.pow(array.get(p, i) - p_avg, 2)

r = -math.log10(1 - r1 / r2)

//get regression standard devition

_s = array.new_float(array_size, 0)

for i = 0 to array_size - 1

array.set(_s, i, array.get(p, i) - array.get(_p, i))

s = array.stdev(_s)

//return results

[a, b, r, s]

// @function top_trend_line_lsr: Trend line fitting based on least square algorithm

// @param series int[] t: time scale value array corresponding to price

// @param series float[] p: price scale value array corresponding to time

// @param series int array_size: the length of regression array

// @param string reg_type: regression type in 'top' and 'bottom'

// @param series int max_iter: maximum fitting iterations

// @param series int min_points: the threshold of regression point numbers

// @returns reg_slop, reg_intercept, reg_level, reg_stdev, reg_point_num

export trend_line_lsr(series int[] t, series float[] p, series int array_size, string reg_type, series int max_iter, series int min_points) =>

[a, b, r, s] = basic_lsr(t, p, array_size)

x1 = array.get(t, 0)

x2 = array.get(t, array_size - 1)

n = array_size //Function arguments cannot be mutable, what the fuck!

for i = 0 to max_iter - 1

//exit conditions

if n < min_points

break

//init new tick and price array

_t = array.new_int(n, 0)

_p = array.new_float(n, 0)

_n = 0

//fine the ground truth values that bigger than predictions

for j = 0 to n - 1

if reg_type == 'top'

if (a * array.get(t, j) + b) < array.get(p, j)

array.set(_t, _n, array.get(t, j))

array.set(_p, _n, array.get(p, j))

_n += 1

else if reg_type == 'bottom'

if (a * array.get(t, j) + b) > array.get(p, j)

array.set(_t, _n, array.get(t, j))

array.set(_p, _n, array.get(p, j))

_n += 1

else

break

//exit if new array size is less than threshold

if _n < min_points

break

//override result if r of new array is bigger than last array

[_a, _b, _r, _s] = basic_lsr(_t, _p, _n)

if _r > r

x1 := array.get(_t, 0)

x2 := array.get(_t, _n - 1)

a := _a

b := _b

r := _r

s := _s

n := _n

//after for loop

[x1, x2, a, b, r, s, n]

Now below is indicator, above was using the lib.

// This source code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/
// © dandrideng

//@version=5
indicator(title="Trend Line Regression", shorttitle="TLR", overlay=true, max_bars_back=800, max_lines_count=400, max_labels_count=400)

import dandrideng/least_squares_regression/1 as lsr

//trend line regression
lookback_left = input.int(defval=5, title="Pivot Lookback Left", group="Trend Line Regression")
lookback_right = input.int(defval=5, title="Pivot Lookback Right", group="Trend Line Regression")
reg_forward = input.int(defval=20, title="Max Lookback Forward", group="Trend Line Regression")
min_length = input.int(defval=100, title="Min Regression Length", maxval=500, group="Trend Line Regression")
max_length = input.int(defval=200, title="Max Regression Length", maxval=500, group="Trend Line Regression")
length_step = input.int(defval=100, title="Regression Length Steps", group="Trend Line Regression")
max_iteration = input.int(defval=4, title="Max Iterations of Regression", group="Trend Line Regression")
max_points = input.int(defval=30, title="Max Point Numbers of Trend Line", group="Trend Line Regression")
min_points = input.int(defval=3, title="Min Point Numbers of Trend Line", group="Trend Line Regression")
min_reg_level = input.float(defval=4, title="Min Regression Level of Trend Line", group="Trend Line Regression")
std_offset = input.float(defval=0.5, title="Multiply Regression Std", group="Trend Line Regression")
top_line_color = input.color(defval=color.red, title="Color of Top Trend Line", group="Trend Line Regression")
bottom_line_color = input.color(defval=color.green, title="Color of Bottom Trend Line", group="Trend Line Regression")
extend_lines = input.bool(defval=true, title="Extend Lines", group="Trend Line Regression") ? extend.right : extend.none
draw_pivot = input.bool(defval=false, title="Draw Piovt High/Low", group="Trend Line Regression")
show_labels = input.bool(defval=false, title="Show Labels", group="Trend Line Regression")

pivot_high = ta.pivothigh(high, lookback_left, lookback_right)
pivot_low = ta.pivotlow(low, lookback_left, lookback_right)

plot(draw_pivot ? pivot_high : na, offset=-lookback_left)
plot(draw_pivot ? pivot_low : na, offset=-lookback_left)

//check regression (functional)
check_trend_line(start, end, type, a, b) =>
    res = true
    for i = start + 1 to end - 1
        yi = a * i + b
        yi_1 = a * (i - 1) + b
        if type == "top" and (close[i] > yi and close[i - 1] > yi_1)
            res := false
            break
        else if type == "bottom" and (close[i] < yi and close[i - 1] < yi_1)
            res := false
            break
    
    res
    
//draw trend line regressions (functional)
draw_trend_line(forward, length, type) =>
    pivot_src = type == "top" ? pivot_high : pivot_low

    tick = array.new_int(max_points, 0)
    price = array.new_float(max_points, 0)
    num = 0

    for i = 0 to length - 1
        if not na(pivot_src[i + forward]) and num < max_points
            array.set(tick, num, i + lookback_left + forward)
            array.set(price, num, pivot_src[i + forward])
            num := num + 1
        
    if num > 0
        [x2, x1, a, b, r, s, n] = lsr.trend_line_lsr(tick, price, num, type, max_iteration, min_points)
        y1 = x1 * a + b
        y2 = x2 * a + b
        o = s * std_offset
        
        if n > min_points and r > min_reg_level and check_trend_line(x1, x2, type, a, b)
            var line mid_line = na
            mid_line := line.new(x1=bar_index-x1, y1=y1, x2=bar_index-x2, y2=y2)
            line.set_width(mid_line, 1)
            line.set_color(mid_line, color.new(type == "top" ? top_line_color : bottom_line_color, 25))
            line.set_extend(mid_line, extend_lines)
            line.set_style(mid_line, line.style_dashed)
        
            var line top_line = na
            top_line := line.new(x1=bar_index-x1, y1=y1+o, x2=bar_index-x2, y2=y2+o)
            line.set_width(top_line, 1)
            line.set_color(top_line, color.new(type == "top" ? top_line_color : bottom_line_color, 25))
            line.set_extend(top_line, extend_lines)
            
            var line bottom_line = na
            bottom_line := line.new(x1=bar_index-x1, y1=y1-o, x2=bar_index-x2, y2=y2-o)
            line.set_width(bottom_line, 1)
            line.set_color(bottom_line, color.new(type == "top" ? top_line_color : bottom_line_color, 25))
            line.set_extend(bottom_line, extend_lines)
            
            linefill.new(mid_line, top_line, color=color.new(type == "top" ? top_line_color : bottom_line_color, 90))
            linefill.new(mid_line, bottom_line, color=color.new(type == "top" ? top_line_color : bottom_line_color, 90))
            
            if show_labels
                var label reg_label = na
                reg_label := label.new(x=bar_index-x2, y=type == "top" ? y2+o : y2-o, textcolor=color.white)
                label.set_color(reg_label, color.new(type == "top" ? top_line_color : bottom_line_color, 30))
                label.set_text(reg_label, "Level: " + str.tostring(r, "#.###") + "\nPoints: " + str.tostring(n, "#"))
                label.set_style(reg_label, type == "top" ? label.style_label_down : label.style_label_up)

//delete all lines
all_lines = line.all
if array.size(all_lines) > 0
	for i = 0 to array.size(all_lines) - 1
		line.delete(array.get(all_lines, i))
		
//delete all labels
all_labels = label.all
if array.size(all_labels) > 0
	for i = 0 to array.size(all_labels) - 1
		label.delete(array.get(all_labels, i))

//trend line
for l = min_length to max_length by length_step
    draw_trend_line(reg_forward, l, "top")
    draw_trend_line(reg_forward, l, "bottom")
        
//end of file

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

// This source code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/

// © dandrideng

//@version=5

indicator(title="Trend Line Regression", shorttitle="TLR", overlay=true, max_bars_back=800, max_lines_count=400, max_labels_count=400)

import dandrideng/least_squares_regression/1 as lsr

//trend line regression

lookback_left = input.int(defval=5, title="Pivot Lookback Left", group="Trend Line Regression")

lookback_right = input.int(defval=5, title="Pivot Lookback Right", group="Trend Line Regression")

reg_forward = input.int(defval=20, title="Max Lookback Forward", group="Trend Line Regression")

min_length = input.int(defval=100, title="Min Regression Length", maxval=500, group="Trend Line Regression")

max_length = input.int(defval=200, title="Max Regression Length", maxval=500, group="Trend Line Regression")

length_step = input.int(defval=100, title="Regression Length Steps", group="Trend Line Regression")

max_iteration = input.int(defval=4, title="Max Iterations of Regression", group="Trend Line Regression")

max_points = input.int(defval=30, title="Max Point Numbers of Trend Line", group="Trend Line Regression")

min_points = input.int(defval=3, title="Min Point Numbers of Trend Line", group="Trend Line Regression")

min_reg_level = input.float(defval=4, title="Min Regression Level of Trend Line", group="Trend Line Regression")

std_offset = input.float(defval=0.5, title="Multiply Regression Std", group="Trend Line Regression")

top_line_color = input.color(defval=color.red, title="Color of Top Trend Line", group="Trend Line Regression")

bottom_line_color = input.color(defval=color.green, title="Color of Bottom Trend Line", group="Trend Line Regression")

extend_lines = input.bool(defval=true, title="Extend Lines", group="Trend Line Regression") ? extend.right : extend.none

draw_pivot = input.bool(defval=false, title="Draw Piovt High/Low", group="Trend Line Regression")

show_labels = input.bool(defval=false, title="Show Labels", group="Trend Line Regression")

pivot_high = ta.pivothigh(high, lookback_left, lookback_right)

pivot_low = ta.pivotlow(low, lookback_left, lookback_right)

plot(draw_pivot ? pivot_high : na, offset=-lookback_left)

plot(draw_pivot ? pivot_low : na, offset=-lookback_left)

//check regression (functional)

check_trend_line(start, end, type, a, b) =>

res = true

for i = start + 1 to end - 1

yi = a * i + b

yi_1 = a * (i - 1) + b

if type == "top" and (close[i] > yi and close[i - 1] > yi_1)

res := false

break

else if type == "bottom" and (close[i] < yi and close[i - 1] < yi_1)

res := false

break

res

//draw trend line regressions (functional)

draw_trend_line(forward, length, type) =>

pivot_src = type == "top" ? pivot_high : pivot_low

tick = array.new_int(max_points, 0)

price = array.new_float(max_points, 0)

num = 0

for i = 0 to length - 1

if not na(pivot_src[i + forward]) and num < max_points

array.set(tick, num, i + lookback_left + forward)

array.set(price, num, pivot_src[i + forward])

num := num + 1

if num > 0

[x2, x1, a, b, r, s, n] = lsr.trend_line_lsr(tick, price, num, type, max_iteration, min_points)

y1 = x1 * a + b

y2 = x2 * a + b

o = s * std_offset

if n > min_points and r > min_reg_level and check_trend_line(x1, x2, type, a, b)

var line mid_line = na

mid_line := line.new(x1=bar_index-x1, y1=y1, x2=bar_index-x2, y2=y2)

line.set_width(mid_line, 1)

line.set_color(mid_line, color.new(type == "top" ? top_line_color : bottom_line_color, 25))

line.set_extend(mid_line, extend_lines)

line.set_style(mid_line, line.style_dashed)

var line top_line = na

top_line := line.new(x1=bar_index-x1, y1=y1+o, x2=bar_index-x2, y2=y2+o)

line.set_width(top_line, 1)

line.set_color(top_line, color.new(type == "top" ? top_line_color : bottom_line_color, 25))

line.set_extend(top_line, extend_lines)

var line bottom_line = na

bottom_line := line.new(x1=bar_index-x1, y1=y1-o, x2=bar_index-x2, y2=y2-o)

line.set_width(bottom_line, 1)

line.set_color(bottom_line, color.new(type == "top" ? top_line_color : bottom_line_color, 25))

line.set_extend(bottom_line, extend_lines)

linefill.new(mid_line, top_line, color=color.new(type == "top" ? top_line_color : bottom_line_color, 90))

linefill.new(mid_line, bottom_line, color=color.new(type == "top" ? top_line_color : bottom_line_color, 90))

if show_labels

var label reg_label = na

reg_label := label.new(x=bar_index-x2, y=type == "top" ? y2+o : y2-o, textcolor=color.white)

label.set_color(reg_label, color.new(type == "top" ? top_line_color : bottom_line_color, 30))

label.set_text(reg_label, "Level: " + str.tostring(r, "#.###") + "\nPoints: " + str.tostring(n, "#"))

label.set_style(reg_label, type == "top" ? label.style_label_down : label.style_label_up)

//delete all lines

all_lines = line.all

if array.size(all_lines) > 0

for i = 0 to array.size(all_lines) - 1

line.delete(array.get(all_lines, i))

//delete all labels

all_labels = label.all

if array.size(all_labels) > 0

for i = 0 to array.size(all_labels) - 1

label.delete(array.get(all_labels, i))

//trend line

for l = min_length to max_length by length_step

draw_trend_line(reg_forward, l, "top")

draw_trend_line(reg_forward, l, "bottom")

//end of file

ashehzi

Participant

Topics: 29

Replies: 88

Been thanked: 9 times

10/30/2023 at 11:31 AM #223047

JC_Bywan

Hi, I’ve added it to the list of conversions to assess. If someone “speaks” the TV language and wants to attempt the conversion before Nicolas is back, please feel free to proceed, thanks. (Edit: I didn’t phrase this very well, any translation contribution is welcome at any time, whether he’s here or not)

JC_Bywan

Moderator

Topics: 20

Replies: 2657

Been thanked: 925 times

01/22/2024 at 11:32 PM #226740

ashehzi

I feel it can be a good trading strat if converted to PRT.

ashehzi

Participant

Topics: 29

Replies: 88

Been thanked: 9 times

01/24/2024 at 9:29 AM #226820

LucasBest

As this indicator use a lot of functions, it would be more easy to understand its spirit (how it works to determine supports and resistance) and then code it in PRT language than trying to convert it like it have been coded in Tradingview language.

Also note that the use of linear regression is not mandatory here. I have done alsmost the same without using it.

LucasBest

Participant

Topics: 16

Replies: 232

Been thanked: 84 times

Author

Posts

Viewing 4 posts - 1 through 4 (of 4 total)

Find exclusive trading pro-tools on