Search code examples
rforecastingtidyverse

R - Date assignation issue - How to correct it?

I have an issue trying to assign a value. This looks strange because after I apply the same commands to another dataset, it works.

Below you will find some screenshots:

Dataset #1 Step 1: Load dataset 1 and use ts function. Dataset #1

Step 2: Create a new object, based on dataset 1 and reformat the date. (Created wrongly) Reshape date values

Dataset #2 Step 1: Load dataset 1 and use ts function. Load dataset #2

Step 2: Create a new object, based on dataset 1 and reformat the date. (Created correctly) enter image description here

Here is the code shown above for the correct code:

# Load required libraries
library(forecast)
library(lubridate)
library(tidyverse)
library(scales)
library(ggfortify)

# Load historical SLA definitions
bwi <- read.csv(file="C:/Users/nsoria/Documents/Data Science/SLA Prediction/TEC_BWI.csv", header=TRUE, sep=';', dec=",")

# Create time series object
ts_bwi <- ts(bwi$SLA, frequency = 12, start = c(2015,1))

############################################ STL Model Algorithm
# Pull out the seasonal, trend, and irregular components from the time series
model_stl <- stl(ts_bwi, s.window = "periodic")

############################################ ARIMA Model Algorithm
# Pull out the seasonal, trend, and irregular components from the time series
#model_arima <- auto.arima(ts_bwi)

# Predict the next 5 month of SLA (STL)
pred <- forecast(model_stl, h = 5)

# Predict the next 5 month of SLA (ARIMA)
#pred <- forecast(model_arima, h = 5)

# Convert pred from list to data frame object
df1 <- fortify(pred) %>% as_tibble()

# Convert ts decimal time to Date class
df1$Date <- as.Date(date_decimal(df1$Index), "%Y-%m-%d")

Here is the link for the dataset that is failing.

Thank you. Nicolás.

Solution

  • It's doing what you ask it to do.

    The "wrong" df1$Index is already a date. You don't need to convert it to anything. When you do, espicially they way you did, funny things happen.

    The date_decimal function wants a number like 2015.123 and gives you "2015-02-14 21:28:48 UTC".

    If you do something crazy like:

    date_decimal(as.Date("2015-01-01"))

    then it first treats as.Date("2015-01-01") as the number it actually is (16436, i.e. the number of days since 1970-01-01) and then it gives you 16436-01-01, (i.e. exactly what you asked for.)

    Your solution is to notice you already have a date, so there's nothing to convert into date.

    Also, your code is giving you warnings. Don't ignore warnings.