Machine learning in trading: theory, models, practice and algo-trading - page 1932

[mytarmailS]

mytarmailS:

no package is needed, where is it written?

Non-finite entries in the input matrix>>> Non-finite entries in the input matrix

maybe there is an inf

You messed up something with the data. Maybe you should try to work out a simple syntax and then train your models.

Try to apply this to the data before you train the model

X <- as.matrix(X)
X[is.na(X)] <- 0
X[is.infinite(X)] <- 100

X is a matrix with predictors

Maxim Dmitrievsky:

5k views, ala village

any luck with your own thing, city? )

[Maxim Dmitrievsky]

mytarmailS:

Try to apply this to the data before training the model

X - matrix with predictors

Did you manage to complete your device, city? )

No, no one helps, only nudges

[mytarmailS]

Who knows at least something about polynomial or harmonic approximation PLEASE!!!

[Aleksey Vyazmikin]

mytarmailS:

no package is needed, where is it written?

Non-finite entries in the input matrix>>> Non-finite entries in the input matrix

maybe there is an inf

You screwed up something with the data. Maybe you should try to understand elementary syntax and only after that you should train your models.

What I messed up - I gave the column with the closing prices

way <- "F:\\FX\\Открытие Брокер_Demo\\Musor\\000\\Save_OHLC.csv"  #  ваш путь

dt <- read.csv(file = way,header = T,sep = ";") #  читаем файл

clos <- dt$Close #  цеНу закрытия в отдельную переменную


get.ind <- function(x,n=5){

I am attaching the data, help me understand, please.

[mytarmailS]

Aleksey Vyazmikin:

What did I do wrong - gave the column with closing prices

I must have screwed up(( type in the commands I gave you

[Aleksey Vyazmikin]

mytarmailS:

I must have screwed up something(( type in the commands I gave you.

Error in x2set(Xsub, n_neighbors, metric, nn_method = nn_sub, n_trees,  : 
  Non-finite entries in the input matrix

But, maybe I put it in the wrong place?

get.target <- function(x, change){
  zz <- TTR::ZigZag(x,change = change,percent = F)
  zz <- c(diff(zz),0) ; zz[zz>=0] <- 1 ; zz[zz<0] <- -1
  return(zz)
}

X <- as.matrix(X)
X[is.na(X)] <- 0
X[is.infinite(X)] <- 100

X <- get.ind(clos)
Y <- as.factor(get.target(clos,change = 0.001))


library(uwot)

What is the 0.001 - what is the meaning of this?

[mytarmailS]

Aleksey Vyazmikin:

But maybe I put it in the wrong place?

No, you didn't))

You want to manipulate "X" before it was created

X <- get.ind(clos)
X <- as.matrix(X)
X[is.na(X)] <- 0

do it like this.

Aleksey Vyazmikin:

What's 0.001 - what's the meaning?

the size of the knee in points or something, depending on the number of digits

you do not know what is what, look :

get.target <- function(x, change){
  zz <- TTR::ZigZag(x,change = change,percent = F)
  zz <- c(diff(zz),0) ; zz[zz>=0] <- 1 ; zz[zz<0] <- -1
  return(zz)
}

we see that this is a TTR packet.

write it in the console.

?TTR::ZigZag

get help

igZag {TTR}     R Documentation
Zig Zag
Description
Zig Zag higlights trends by removing price changes smaller than change and interpolating lines between the extreme points.

Usage
ZigZag(HL, change = 10, percent = TRUE, retrace = FALSE,
  lastExtreme = TRUE)
Arguments
HL      
Object that is coercible to xts or matrix and contains either a High-Low price series, or a Close price series.

change  
Minimum price movement, either in dollars or percent (see percent).

percent 
Use percentage or dollar change?

retrace 
Is change a retracement of the previous move, or an absolute change from peak to trough?

lastExtreme     
If the extreme price is the same over multiple periods, should the extreme price be the first or last observation?

Details
The Zig Zag is non-predictive. The purpose of the Zig Zag is filter noise and make chart patterns clearer. It's more a visual tool than an indicator.

Value
A object of the same class as HL or a vector (if try.xts fails) containing the Zig Zag indicator.

Warning
The last value of the ZigZag indicator is unstable (i.e. unknown) until the turning point actually occurs. Therefore this indicator isn't well-suited for use for systematic trading strategies.

Note
If High-Low prices are given, the function calculates the max/min using the high/low prices. Otherwise the function calculates the max/min of the single series.

Author(s)
Joshua Ulrich

References
The following site(s) were used to code/document this indicator:
http://www.fmlabs.com/reference/default.htm?url=ZigZag.htm
https://www.linnsoft.com/techind/zig-zag-indicator-zig-zzo
https://www.linnsoft.com/techind/zig-zag-oscillator-indicator-zzo
http://www.metastock.com/Customer/Resources/TAAZ/#127
http://www.stockcharts.com/school/doku.php?id=chart_school:technical_indicators:zigzag
Examples

## Get Data and Indicator ##
data(ttrc)
zz <- ZigZag( ttrc[,c("High", "Low")], change=20 )

[Aleksey Vyazmikin]

mytarmailS:

Not there ))

You want to manipulate "X" before it was created

do it like this

the size of the X knee in points or whatever, depending on the number of digits.

you don't know what is what, look :

we see that this is a TTR packet.

write it in the console.

get help.

Yeah, the next thing was rustling around and I got this

> X <- as.matrix(X)

> X[is.na(X)] <- 0

> X[is.infinite(X)] <- 100

> library(uwot)

> train.idx <- 100:8000

> test.idx <- 8001:10000

> UM <- umap(X = X[train.idx,],
+            y = Y[train.idx], 
+            approx_pow = TRUE, 
+            n_components = 3, 
+            ret_mode .... [TRUNCATED] 

> predict.train <- umap_transform(X = X[train.idx,], 
+                                 model = UM, n_threads = 4 L, 
+                                 .... [TRUNCATED] 
11:08:52 Read 7901 rows and found 31 numeric columns
11:08:52 Applying training data column filtering/scaling
11:08:52 Processing block 1 of 1
11:08:52 Writing NN index file to temp file C:\Users\S_V_A\AppData\Local\Temp\RtmpK0sSR3\file28f467fe2995
11:08:52 Searching Annoy index using 4 threads, search_k = 1500
11:08:52 Commencing smooth kNN distance calibration using 4 threads
11:08:52 Initializing by weighted average of neighbor coordinates using 4 threads
11:08:52 Commencing optimization for 167 epochs, with 118515 positive edges
0%   10   20   30   40   50   60   70   80   90   100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
11:08:54 Finished

> predict.test <- umap_transform(X = X[test.idx,], 
+                                model = UM, n_threads = 4 L, 
+                                ver .... [TRUNCATED] 
11:08:54 Read 2000 rows and found 31 numeric columns
11:08:54 Applying training data column filtering/scaling
11:08:54 Processing block 1 of 1
11:08:54 Writing NN index file to temp file C:\Users\S_V_A\AppData\Local\Temp\RtmpK0sSR3\file28f45e3376d9
11:08:54 Searching Annoy index using 4 threads, search_k = 1500
11:08:55 Commencing smooth kNN distance calibration using 4 threads
11:08:55 Initializing by weighted average of neighbor coordinates using 4 threads
11:08:55 Commencing optimization for 167 epochs, with 30000 positive edges
0%   10   20   30   40   50   60   70   80   90   100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
11:08:55 Finished

> library(car)
Загрузка требуемого пакета: carData

> scatter3d(x = predict.train[,1], 
+           y = predict.train[,2], 
+           z = predict.train[,3],
+           groups = Y[train.idx],
+        .... [TRUNCATED] 
Загрузка требуемого пакета: rgl
Загрузка требуемого пакета: mgcv

If I have a step change price of 1, then I should change 0.001 to e.g. 100?

I changed it to 100.

11:15:15 Read 2000 rows and found 31 numeric columns
11:15:15 Applying training data column filtering/scaling
11:15:15 Processing block 1 of 1
11:15:15 Writing NN index file to temp file C:\Users\S_V_A\AppData\Local\Temp\RtmpK0sSR3\file28f417876e23
11:15:15 Searching Annoy index using 4 threads, search_k = 1500
11:15:15 Commencing smooth kNN distance calibration using 4 threads
11:15:15 Initializing by weighted average of neighbor coordinates using 4 threads
11:15:15 Commencing optimization for 167 epochs, with 30000 positive edges
0%   10   20   30   40   50   60   70   80   90   100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
11:15:16 Finished

> library(car)

> scatter3d(x = predict.train[,1], 
+           y = predict.train[,2], 
+           z = predict.train[,3],
+           groups = Y[train.idx],
+        .... [TRUNCATED] 
> 

[mytarmailS]

Aleksey Vyazmikin:

Yeah, it went on and on, and I got this

If I have a step change price 1, then I need to change 0.001 to for example 100?

Play with change parameter

last200clos <- tail(clos,200)
plot(last200clos,t="l")
zz <- TTR::ZigZag(HL = last200clos , percent = F,change = 50)
lines(zz,col=2,lwd=2)

[Aleksey Vyazmikin]

mytarmailS:

Play with the change parameter

Got this picture from the last results - 4 clear clusters

Earlier you talked about some numerical values to evaluate the results - what numbers do you suggest to look at?