suppressMessages({
library(modeltime)
library(timetk)
library(lubridate)
library(tidyverse)
library(data.table)
library(tidymodels)
library(modeltime.ensemble)
})
options(repr.plot.width = 12, repr.plot.height = 8)
knitr::opts_chunk$set(echo = TRUE)
train <- read.csv("~/Desktop/train.csv")
train$date= as.Date(train$date, "%m/%d/%Y %H:%M")
train=train %>%
group_by(date) %>%
summarise_each(funs(sum))
train %>% head(3)
## # A tibble: 3 x 5
## date user session visitors views
## <date> <int> <int> <int> <int>
## 1 2018-09-09 281 266 73 1826
## 2 2018-09-10 264 247 51 2092
## 3 2018-09-11 329 310 58 1998
Time series Plot
## Using date_var: date
Arima model
## frequency = 7 observations per 1 week
## parsnip model object
##
## Fit time: 1.8s
## Series: outcome
## ARIMA(1,0,2)(0,1,2)[7] with drift
##
## Coefficients:
## ar1 ma1 ma2 sma1 sma2 drift
## 0.8674 -0.2586 -0.1947 -0.8988 0.0589 3.4389
## s.e. 0.0423 0.0599 0.0520 0.0400 0.0370 0.9404
##
## sigma^2 estimated as 66722: log likelihood=-5049.95
## AIC=10113.89 AICc=10114.05 BIC=10145.98
Prophet model
## Disabling daily seasonality. Run prophet with daily.seasonality=TRUE to override this.
## parsnip model object
##
## Fit time: 2.8s
## PROPHET Model
## - growth: 'linear'
## - n.changepoints: 25
## - changepoint.range: 0.8
## - yearly.seasonality: 'TRUE'
## - weekly.seasonality: 'auto'
## - daily.seasonality: 'auto'
## - seasonality.mode: 'additive'
## - changepoint.prior.scale: 0.05
## - seasonality.prior.scale: 10
## - holidays.prior.scale: 10
## - logistic_cap: NULL
## - logistic_floor: NULL
## - extra_regressors: 0
Elastic Net
Sub-Model Evaluation
## # Modeltime Table
## # A tibble: 2 x 3
## .model_id .model .model_desc
## <int> <list> <chr>
## 1 1 <fit[+]> ARIMA(1,0,2)(0,1,2)[7] WITH DRIFT
## 2 2 <fit[+]> PROPHET
| .model_id |
.model_desc |
.type |
mae |
mape |
mase |
smape |
rmse |
rsq |
| 1 |
ARIMA(1,0,2)(0,1,2)[7] WITH DRIFT |
Test |
516.49 |
22.06 |
0.94 |
19.51 |
683.77 |
0.42 |
| 2 |
PROPHET |
Test |
686.46 |
31.91 |
1.25 |
26.86 |
817.22 |
0.54 |
## Warning: Expecting the following names to be in the data frame: .conf_hi, .conf_lo.
## Proceeding with '.conf_interval_show = FALSE' to visualize the forecast without confidence intervals.
## Alternatively, try using `modeltime_calibrate()` before forecasting to add confidence intervals.
## ── Modeltime Ensemble ───────────────────────────────────────────
## Ensemble of 2 Models (MEAN)
##
## # Modeltime Table
## # A tibble: 2 x 3
## .model_id .model .model_desc
## <int> <list> <chr>
## 1 1 <fit[+]> ARIMA(1,0,2)(0,1,2)[7] WITH DRIFT
## 2 2 <fit[+]> PROPHET
## ── Modeltime Ensemble ───────────────────────────────────────────
## Ensemble of 2 Models (MEAN)
##
## # Modeltime Table
## # A tibble: 2 x 3
## .model_id .model .model_desc
## <int> <list> <chr>
## 1 1 <fit[+]> ARIMA(1,0,2)(0,1,2)[7] WITH DRIFT
## 2 2 <fit[+]> PROPHET
## # Modeltime Table
## # A tibble: 2 x 3
## .model_id .model .model_desc
## <int> <list> <chr>
## 1 1 <ensemble [2]> ENSEMBLE (MEAN): 2 MODELS
## 2 2 <ensemble [2]> ENSEMBLE (MEDIAN): 2 MODELS
| .model_id |
.model_desc |
.type |
mae |
mape |
mase |
smape |
rmse |
rsq |
| 1 |
ENSEMBLE (MEAN): 2 MODELS |
Test |
583.7 |
26.24 |
1.07 |
22.78 |
726.15 |
0.46 |
| 2 |
ENSEMBLE (MEDIAN): 2 MODELS |
Test |
583.7 |
26.24 |
1.07 |
22.78 |
726.15 |
0.46 |
ensemble_models_tbl %>%
modeltime_forecast(
new_data = testing(splits),
actual_data = train
) %>%
plot_modeltime_forecast(.interactive =FALSE)
## Warning: Expecting the following names to be in the data frame: .conf_hi, .conf_lo.
## Proceeding with '.conf_interval_show = FALSE' to visualize the forecast without confidence intervals.
## Alternatively, try using `modeltime_calibrate()` before forecasting to add confidence intervals.
ensemble_models_tbl %>%
modeltime_forecast(
new_data = testing(splits),
actual_data = train
) %>%
plot_modeltime_forecast(.interactive =FALSE)
## Warning: Expecting the following names to be in the data frame: .conf_hi, .conf_lo.
## Proceeding with '.conf_interval_show = FALSE' to visualize the forecast without confidence intervals.
## Alternatively, try using `modeltime_calibrate()` before forecasting to add confidence intervals.
Predict next two month
ensemble_models_tbl %>%
# Remove ARIMA model with low accuracy
filter(.model_id != 3) %>%
# Refit and Forecast Forward
modeltime_refit(train) %>%
modeltime_forecast(h = "2 months", actual_data = train) %>%
plot_modeltime_forecast(.interactive = FALSE)
## frequency = 7 observations per 1 week
## Disabling daily seasonality. Run prophet with daily.seasonality=TRUE to override this.
## frequency = 7 observations per 1 week
## Disabling daily seasonality. Run prophet with daily.seasonality=TRUE to override this.
## Warning: Expecting the following names to be in the data frame: .conf_hi, .conf_lo.
## Proceeding with '.conf_interval_show = FALSE' to visualize the forecast without confidence intervals.
## Alternatively, try using `modeltime_calibrate()` before forecasting to add confidence intervals.
predict<-ensemble_models_tbl %>%
# Remove ARIMA model with low accuracy
filter(.model_id != 3) %>%
# Refit and Forecast Forward
modeltime_refit(train) %>%
modeltime_forecast(h = "2 months", actual_data = train)
## frequency = 7 observations per 1 week
## Disabling daily seasonality. Run prophet with daily.seasonality=TRUE to override this.
## frequency = 7 observations per 1 week
## Disabling daily seasonality. Run prophet with daily.seasonality=TRUE to override this.
write.csv(predict,"~/Desktop/predict.csv", row.names = FALSE)