Basis van tidy machine learning

Machine Learning in de tidyverse

Dmitriy (Dima) Gorenshteyn

Lead Data Scientist, Memorial Sloan Kettering Cancer Center

De kern van tidy machine learning

Workflow met lijstkolommen

De Gapminder-dataset

pakket dslabs
Observaties: 77 landen, 52 jaar per land (1960-2011)
Features:
- year
- infant_mortality
- life_expectancy
- fertility
- population
- gdpPercap

Workflow met lijstkolommen

Stap 1: Maak een lijstkolom - nest je data

Nesten per land

library(tidyverse)
nested <- gapminder %>%
          group_by(country) %>%
          nest()

Een geneste tibble bekijken

> nested$data[[4]]
# A tibble: 52 x 6
    year infant_mortality life_expectancy fertility population gdpPercap
   <int>            <dbl>           <dbl>     <dbl>      <dbl>     <int>
 1  1960             37.3            68.8      2.70    7065525      7415
 2  1961             35.0            69.7      2.79    7105654      7781
 3  1962             32.9            69.5      2.80    7151077      7937
 4  1963             31.2            69.6      2.82    7199962      8209
 5  1964             29.7            70.1      2.80    7249855      8652
 6  1965             28.3            69.9      2.70    7298794      8893

Stap 3: Lijstkolommen vereenvoudigen - unnest()

nested %>% 
  unnest(data)

# A tibble: 4,004 x 7
   country  year infant_mortality life_expectancy fertility population   ...
   <fct>   <int>            <dbl>           <dbl>     <dbl>      <dbl>   ...
 1 Algeria  1960              148            47.5      7.65   11124892   ...
 2 Algeria  1961              148            48.0      7.65   11404859   ...
 3 Algeria  1962              148            48.6      7.65   11690152   ...
 4 Algeria  1963              148            49.1      7.65   11985130   ...
 5 Algeria  1964              149            49.6      7.65   12295973   ...
 6 Algeria  1965              149            50.1      7.66   12626953   ...

Laten we beginnen!

Machine Learning in de tidyverse