Lineárna regresia

1. Načítanie dát

Dataset: DailyDelhiClimateTest.csv

data_path <- "DailyDelhiClimateTest.csv"
df <- read.csv(data_path, stringsAsFactors = FALSE)
str(df)

'data.frame':   114 obs. of  5 variables:
 $ date        : chr  "2017-01-01" "2017-01-02" "2017-01-03" "2017-01-04" ...
 $ meantemp    : num  15.9 18.5 17.1 18.7 18.4 ...
 $ humidity    : num  85.9 77.2 81.9 70 74.9 ...
 $ wind_speed  : num  2.74 2.89 4.02 4.54 3.3 ...
 $ meanpressure: num  59 1018 1018 1016 1014 ...

head(df)

2. Predspracovanie

• Skontrolujeme chýbajúce hodnoty a typy stĺpcov.
• Vytvoríme rok, mesiac a deň-of-week premenné.

# Prehľad chýbajúcich hodnôt
map_df(df, ~ sum(is.na(.))) %>% gather(key='variable', value='missing') %>% arrange(desc(missing))

# Convert to tidy date parts
df <- df %>% mutate(year = year(date), month = month(date, label = TRUE, abbr = TRUE),
                    day = day(date), weekday = wday(date, label = TRUE))

3. Exploratory data analysis (EDA)

3.1 Trendy pri teplote (avg temp)

df %>% ggplot(aes(x=date, y=meantemp)) +
  geom_line() +
  labs(title='Denný priemer teploty v Dillí (test set)', x='Dátum', y='Mean Temperature (°C)') +
  theme_minimal()

3.2 Sezónnosť podľa mesiaca

df %>% group_by(month) %>% summarize(mean_meantemp = mean(meantemp, na.rm=TRUE),
                                     sd_meantemp = sd(meantemp, na.rm=TRUE),
                                     n = n()) %>%
  ggplot(aes(x=month, y=mean_meantemp)) +
  geom_col() + labs(title='Priemerná mean teplota podľa mesiaca', x='Mesiac', y='Priemer (°C)') +
  theme_minimal()

3.3 Závislosti medzi premennými

library(GGally)
numeric_cols <- df %>% select_if(is.numeric) %>% select(-year)
GGally::ggpairs(numeric_cols)

4. Jednoduché modelovanie

4.1 Lineárny model — predikcia meantemp z vlhkosti a vetra

lm1 <- lm(meantemp ~ humidity + wind_speed + pressure, data = df)

Error in model.frame.default(formula = meantemp ~ humidity + wind_speed +  : 
  invalid type (list) for variable 'pressure'

4.2 Časová rada — dekompozícia meantemp

5. Krátke závery

Bonus: uloženie upraveného datasetu

---