Dev kzo

[vlamai]

test pull requests. Add Abs function to float and int series

[rocketlaunchr-cto]

Based on your PR, I can see that your intention is to transform all the values in the series to the Absolute value.

The way to do it is actually like this:

s1 := dataframe.NewSeriesInt64("day", nil, 1, 2, 3, 4, 5, 6, 7, 8)

s1.Lock()
n := s1.NRows(Options{DontLock:true})
for row := 0; row < n; row++ {
    val := s1.Value(row, Options{DontLock:true})
    if val == nil { continue }
    s1.Update(row, abs(val), Options{DontLock:true})
}
s1.Unlock()

You can make this into a function of your own.

My opinion is that having an Abs() function for a series doesn't make sense to many types of data types such as strings and dates and perhaps may more. It really only makes sense for numbers.

Do you have any good reasons for including it into the API? Maybe you have some good reasons.

[vlamai]

1. Pandas series have this function.
2. Any function can be done by yourself.
3. It is simplest function that i can find. To test PR and find out what you think about API.

How do you think need try to add all this functions ? pandas Or you have some priorities to add functions ?

[rocketlaunchr-cto]

I can see from the pandas API: http://pandas.pydata.org/pandas-docs/stable/api.html#computations-descriptive-stats

• Abs function
• Max, Min, Sum etc (also asked by https://github.com/rocketlaunchr/dataframe-go/issues/2#issuecomment-435442489)

I think since these are only applicable to SeriesInt64 and SeriesFloat64.

• Maybe create a new sub-package (called stats)
• In the sub-package, create a new interface that embeds the Series interface but adds in all the functions you want to add such as Abs.
• Create a new struct called SeriesInt64 and SeriesFloat64 which embeds the original SeriesInt64

This will allow the main package to contain only the bare essentials.

What do you think of this idea?

[rocketlaunchr-cto]

I also want to add to the main package: Filter, Map and Reduce and use https://github.com/robpike/filter for inspiration.

This is created by the co-creator of Go and can work with any type (including time etc)

[rocketlaunchr-cto]

https://godoc.org/gonum.org/v1/gonum/stat can be leveraged for statistics.

[vlamai]

I'm obviously doing something wrong. create new package stat |-stat |--stat.go

package stat

import (
    "github.com/rocketlaunchr/dataframe-go"
    "math"
)

type Stat interface {
    dataframe.Series
    Abs()
}

type SeriesFloat64 struct {
    dataframe.SeriesFloat64
}

type SeriesInt64 struct {
    dataframe.SeriesInt64
}

func (s *SeriesFloat64) Abs() {
    for key, value := range s.Values {
        val := math.Abs(*value)
        s.Values[key] = &val
    }
}

func (s *SeriesInt64) Abs() {
    for key, value := range s.Values {
        val := *value
        if val < 0 {
            val = val * -1
        }
        s.Values[key] = &val
    }
}

but in stat_test.go i can't call function

package stat

import (
    "github.com/rocketlaunchr/dataframe-go"
    "testing"
)

func TestSeriesFloat64_Abs(t *testing.T) {
    s1 := dataframe.NewSeriesFloat64("num",nil,-1.0,1.0)
    s1.Abs() // error
}

can you create a sample ?

[rocketlaunchr-cto]

You are very close.

The error is because s1 doesn't have the Abs method. s1 is from the dataframe package.

It'll be something like this:

s1 := SeriesFloat64{ dataframe.NewSeriesFloat64("num",nil,-1.0,1.0) }
s1.Abs()

Also remember to check for nil values in your Abs( ) function.

You will also need to lock the series before you manipulate any of the contained values. (since your intention is to do an InPlace manipulation).

[rocketlaunchr-cto]

Something like this:

func (s *SeriesFloat64) Abs() {
    s.Lock()
    defer s.Unlock()

    for key := range s.Values {
        val := s.Values[key]
        if val != nil && *val < 0 {
            s.Values[key] = &[]float64{-1*val}[0]
        }
    }
}

[vlamai]

1. In test it works, in code no.
2. I think this method of using the function is inconvenient

If it's hard make an example of a function. My knowledge is apparently not enough

[rocketlaunchr-cto]

What are you trying to achieve?

[rocketlaunchr-cto]

Also embed the pointer:

type SeriesFloat64 struct {
    *dataframe.SeriesFloat64
}

type SeriesInt64 struct {
    *dataframe.SeriesInt64
}