how to find polygon with given x and y coordinates - plsql

I have data like this, this is coordinates system of (x,y) pairs, so the question is how to get the (x,y) points that if we connect this points other points must be inside of polygon and it must be convex polygon
x_axis y_axis id
8 14.5 1
1.1 1.1 2
3.5 3.4 3
4.5 0.4 4
13.5 7.8 5
11.5 15.2 6
2.8 8.3 7
0.3 5.4 8
1.5 3.8 9
8 8 10
8.3 8.1 11
2 10 12
5 14 13
... ... ...

Naive solution is to take the rectangle that contains all points:
select min(ap.x_axis) as x,
min(ap.y_axis) as y,
'lower_left' as desc
from all_points ap
union all
select min(ap.x_axis) as x,
max(ap.y_axis) as y,
'upper_left' as desc
from all_points ap
union all
select max(ap.x_axis) as x,
max(ap.y_axis) as y,
'upper_right' as desc
from all_points ap
union all
select max(ap.x_axis) as x,
min(ap.y_axis) as y,
'lower_right' as desc
from all_points ap
It is working solution for the problem you posted.
If you want 'the smallest' polygon, this is called 'convex hull'. The algorithm for finding it is called 'Graham scan' and it's descriped on wikipedia.
We won't code it for you.

Related

R: Creating an index vector

I need some help with R coding here.
The data set Glass consists of 214 rows of data in which each row corresponds to a glass sample. Each row consists of 10 columns. When viewed as a classification problem, column 10
(Type) specifies the class of each observation/instance. The remaining columns are attributes that might beused to infer column 10. Here is an example of the first row
RI Na Mg Al Si K Ca Ba Fe Type
1 1.52101 13.64 4.49 1.10 71.78 0.06 8.75 0.0 0.0 1
First, I casted column 10 so that it is interpreted by R as a factor instead of an integer value.
Now I need to create a vector with indices for all observations (must have values 1-214). This needs to be done to creating training data for Naive Bayes. I know how to create a vector with 214 values, but not one that has specific indices for observations from a data frame.
If it helps this is being done to set up training data for Naive Bayes, thanks
I'm not totally sure that I get what you're trying to do... So please forgive me if my solution isn't helpful. If your df's name is 'df', just use the dplyr package for reordering your columns and write
library(dplyr)
df['index'] <- 1:214
df <- df %>% select(index,everything())
Here's an example. So that I can post full dataframes, my dataframes will only have 10 rows...
Let's say my dataframe is:
df <- data.frame(col1 = c(2.3,6.3,9.2,1.7,5.0,8.5,7.9,3.5,2.2,11.5),
col2 = c(1.5,2.8,1.7,3.5,6.0,9.0,12.0,18.0,20.0,25.0))
So it looks like
col1 col2
1 2.3 1.5
2 6.3 2.8
3 9.2 1.7
4 1.7 3.5
5 5.0 6.0
6 8.5 9.0
7 7.9 12.0
8 3.5 18.0
9 2.2 20.0
10 11.5 25.0
If I want to add another column that just is 1,2,3,4,5,6,7,8,9,10... and I'll call it 'index' ...I could do this:
library(dplyr)
df['index'] <- 1:10
df <- df %>% select(index, everything())
That will give me
index col1 col2
1 1 2.3 1.5
2 2 6.3 2.8
3 3 9.2 1.7
4 4 1.7 3.5
5 5 5.0 6.0
6 6 8.5 9.0
7 7 7.9 12.0
8 8 3.5 18.0
9 9 2.2 20.0
10 10 11.5 25.0
Hope this will help
df$ind <- seq.int(nrow(df))

Cumulative summing between groups using dplyr

I have a tibble structured as follows:
day theta
1 1 2.1
2 1 2.1
3 2 3.2
4 2 3.2
5 5 9.5
6 5 9.5
7 5 9.5
Note that the tibble contains multiple rows for each day, and for each day the same value for theta is repeated an arbitrary number of times. (The tibble contains other arbitrary columns necessitating this repeating structure.)
I'd like to use dplyr to cumulatively sum values for theta across days such that, in the example above, 2.1 is added only a single time to 3.2, etc. The tibble would be mutated so as to append the new cumulative sum (c.theta) as follows:
day theta c.theta
1 1 2.1 2.1
2 1 2.1 2.1
3 2 3.2 5.3
4 2 3.2 5.3
5 5 9.5 14.8
6 5 9.5 14.8
7 5 9.5 14.8
...
My initial efforts to group_by day and then cumsum over theta resulted only in cumulative summing over the full set of data (e.g., 2.1 + 2.1 + 3.2 ...) which is undesirable. In my Stack Overflow searches, I can find many examples of cumulative summing within groups, but never between groups, as I describe above. Nudges in the right direction would be much appreciated.
Doing this in dplyr I came up with a very similar solution to PoGibas - use distinct to just get one row per day, find the sum and merge back in:
df = read.table(text="day theta
1 1 2.1
2 1 2.1
3 2 3.2
4 2 3.2
5 5 9.5
6 5 9.5
7 5 9.5", header = TRUE)
cumsums = df %>%
distinct(day, theta) %>%
mutate(ctheta = cumsum(theta))
df %>%
left_join(cumsums %>% select(day, ctheta), by = 'day')
Not a dplyr, but just an alternative data.table solution:
library(data.table)
# Original table is called d
setDT(d)
merge(d, unique(d)[, .(c.theta = cumsum(theta), day)])
day theta c.theta
1: 1 2.1 2.1
2: 1 2.1 2.1
3: 2 3.2 5.3
4: 2 3.2 5.3
5: 5 9.5 14.8
6: 5 9.5 14.8
7: 5 9.5 14.8
PS: If you want to preserve other columns you have to use unique(d[, .(day, theta)])
In base R you could use split<- and tapply to return the desired result.
# construct 0 vector to fill in
dat$temp <- 0
# fill in with cumulative sum for each day
split(dat$temp, dat$day) <- cumsum(tapply(dat$theta, dat$day, head, 1))
Here, tapply returns the first element of theta for each day which is is fed to cumsum. The elements of cumulative sum are fed to each day using split<-.
This returns
dat
day theta temp
1 1 2.1 2.1
2 1 2.1 2.1
3 2 3.2 5.3
4 2 3.2 5.3
5 5 9.5 14.8
6 5 9.5 14.8
7 5 9.5 14.8

How to identify missing rows in a R df - without NA

I have several df with different length of the observations. The observations should have been collected at every 0.2 m depth. This is often the case but sometimes 1, 2 or more depth intervals are "missed out/jumped over", in a completely random order. The tables have same fields. I need to know which depth that has been missed out and at which time. As an example is a reduced file; Table A (profile, time_UTC, depth_m). Table A has 12 rows, but should have 15 and i.e 1 at 0.4m, and at 1.2m and 1.4m. The time stamps (ms) are irregular so I cannot use them to identify gaps in depth.
==============
Table A
Has 12 rows but should have 15:
profil time_UTC depth_m
1 V 24871 0.2
2 V 24877 0.6
3 V 24882 0.8
4 V 24887 1
5 V 24898 1.6
6 V 24901 1.8
7 V 24902 2
8 V 24905 2.2
9 V 24907 2.4
10 V 24909 2.6
11 V 24912 2.8
12 V 24915 3
The check needs to be done in an operational manner for all df that I will read in.
I need help to write a query where I can find those missing rows and add new rows with the missing depth and integrated time.
I provided two links for the similar problem, but in the R example the sequence is categorical (my df are not) and the other is for Sql code.
R identify missing rows from pre-specified sequence
How to find missing rows?
Thanks for the help in advance.
library(dplyr)
right_join(df, data_frame(depth_m = seq(0.2, 3, by = 0.2)))
#> Joining, by = "depth_m"
#> profil time_UTC depth_m
#> 1 V 24871 0.2
#> 2 <NA> NA 0.4
#> 3 <NA> NA 0.6
#> 4 V 24882 0.8
#> 5 V 24887 1.0
#> 6 <NA> NA 1.2
#> 7 <NA> NA 1.4
#> 8 V 24898 1.6
#> 9 V 24901 1.8
#> 10 V 24902 2.0
#> 11 V 24905 2.2
#> 12 <NA> NA 2.4
#> 13 <NA> NA 2.6
#> 14 <NA> NA 2.8
#> 15 V 24915 3.0

Segmenting a data frame by row based on previous rows values

I have a data frame in R that contains 2 columns named x and y (co-ordinates). The data frame represents a journey with each line representing the position at the next point in time.
x y seconds
1 0.0 0.0 0
2 -5.8 -8.5 1
3 -11.6 -18.2 2
4 -16.9 -30.1 3
5 -22.8 -40.8 4
6 -29.0 -51.6 5
I need to break the journey up into segments where each segment starts once the distance from the start of the previous segment crosses a certain threshold (e.g. 200).
I have recently switched from using SAS to R, and this is the first time I've come across anything I can do easily in SAS but can't even think of the way to approach the problem in R.
I've posted the SAS code I would use below to do the same job. It creates a new column called segment.
%let cutoff=200;
data segments;
set journey;
retain segment distance x_start y_start;
if _n_=1 then do;
x_start=x;
y_start=y;
segment=1;
distance=0;
end;
distance + sqrt((x-x_start)**2+(y-y_start)**2);
if distance>&cutoff then do;
x_start=x;
y_start=y;
segment+1;
distance=0;
end;
keep x y seconds segment;
run;
Edit: Example output
If the cutoff were 200 then an example of required output would look something like...
x y seconds segment
1 0.0 0.0 0 1
2 40.0 30.0 1 1
3 80.0 60.0 2 1
4 120.0 90.0 3 1
5 160.0 120.0 4 2
6 120.0 150.0 5 2
7 80.0 180.0 6 2
8 40.0 210.0 7 2
9 0.0 240.0 8 3
If your data set is dd, something like
cutoff <- 200
origin <- dd[1,c("x","y")]
cur.seg <- 1
dd$segment <- NA
for (i in 1:nrow(dd)) {
dist <- sqrt(sum((dd[i,c("x","y")]-origin)^2))
if (dist>cutoff) {
cur.seg <- cur.seg+1
origin <- dd[i,c("x","y")]
}
dd$segment[i] <- cur.seg
}
should work. There are some refinements (it might be more efficient to compute distances of the current origin to all rows, then use which(dist>cutoff)[1] to jump to the first row that goes beyond the cutoff), and it would be interesting to try to come up with a completely vectorized solution, but this should be OK. How big is your data set?

Computing a "rightmost" moving average?

I would like to compute a moving average (ma) over some time series data but I would like the ma to consider the order n starting from the rightmost of my series so my last ma value corresponds to the ma of the last n values of my series. The desired function rightmost_ma would produce this output:
data <- seq(1,10)
> data
[1] 1 2 3 4 5 6 7 8 9 10
rightmost_ma(data, n=2)
NA 1.5 2.5 3.5 4.5 5.5 6.5 7.5 8.5 9.5
I was reviewing the different ma possibilities e.g. package forecast and could not find how to cover this use case. Note that the critical requirement for me is to have valid non NA ma values for the last elements of the series or in other words I want my ma to produce valid results without "looking into the future".
Take a look at rollmean function from zoo package
> library(zoo)
> rollmean(zoo(1:10), 2, align ="right", fill=NA)
1 2 3 4 5 6 7 8 9 10
NA 1.5 2.5 3.5 4.5 5.5 6.5 7.5 8.5 9.5
you can also use rollapply
> rollapply(zoo(1:10), width=2, FUN=mean, align = "right", fill=NA)
1 2 3 4 5 6 7 8 9 10
NA 1.5 2.5 3.5 4.5 5.5 6.5 7.5 8.5 9.5
I think using stats::filter is less complicated, and might have better performance (though zoo is well written).
This:
filter(1:10, c(1,1)/2, sides=1)
gives:
Time Series:
Start = 1
End = 10
Frequency = 1
[1] NA 1.5 2.5 3.5 4.5 5.5 6.5 7.5 8.5 9.5
If you don't want the result to be a ts object, use as.vector on the result.

Resources