Thanks for the explanation! Partial Dependence Calibration are plot give us much better representation than the others.

]]>What you are seeing are the values of the target variable compared to the predicted values (red and blue lines), along the dimension of the independent variable. This lets you see if the values that you are predicting line up with that of the target variable, for all values that the independent variable takes on. A good model will show the two lines very close to each other while a poorly fit model will show a larger difference in the values of lines.

]]>Thanks in Advance.

]]>Do you think, can we use Google Trend(GT) data with CARMA. It has both time trend and calender based variables. Is this right?. I am just thinking how can we put into a group as you did with Sales data of Walmart.

]]>No problem. I designed the CARMA functions to replicate time series models that only utilize features about the target variable, calendar based variables, and time trend variables.

]]>Okey thanks for the response.

]]>The CARMA suite of models are not set up to utilize other predictors. If you want to utilize other predictors, you’ll have to use the Auto_Regression() functions.

]]>Sorry for putting one more question.

I have data of monthly GDP and 25 predictors all data is numeric. As an initial test, I am using your example codes. When using this line of codes XGBoostResults$TimeSeriesPlot, I am not able to get predicted values for in sample (Though, I am getting out of sample predicted).

Please note that I also tested with the single GDP as well by removing all the predictors. But the problem is still there.

I get the warning message,

Warning messages:

1: Removed 10 rows containing missing values (geom_path).

2: Removed 182 rows containing missing values (geom_path).

In your codes for Walmart sales it worked fine. Here are the codes,

XGBoostResults <- RemixAutoML::AutoXGBoostCARMA(

data,

TargetColumnName = "gdpm",

DateColumnName = "date",

GroupVariables = NULL,

FC_Periods = 10,

TimeUnit = "month",

TargetTransformation = TRUE,

Lags = c(1:5),

MA_Periods = c(1:5),

CalendarVariables = TRUE,

HolidayVariable = TRUE,

TimeTrendVariable = TRUE,

DataTruncate = FALSE,

SplitRatios = c(0.70,0.20,0.10),

TreeMethod = "hist",

EvalMetric = "RMSE",

GridTune = FALSE,

GridEvalMetric = "mse",

ModelCount = 1,

NTrees = 5000,

PartitionType = "time",

Timer = TRUE)

XGBoostResults$TimeSeriesPlot

]]>