Skip to contents

Generate 'TimeGPT' forecast

Usage

nixtla_client_forecast(
  df,
  h = 8,
  freq = NULL,
  id_col = NULL,
  time_col = "ds",
  target_col = "y",
  X_df = NULL,
  level = NULL,
  quantiles = NULL,
  finetune_steps = 0,
  finetune_loss = "default",
  clean_ex_first = TRUE,
  add_history = FALSE,
  model = "timegpt-1",
  num_partitions = NULL
)

Arguments

df

A tsibble or a data frame with time series data.

h

Forecast horizon.

freq

Frequency of the data.

id_col

Column that identifies each series.

time_col

Column that identifies each timestep.

target_col

Column that contains the target variable.

X_df

A tsibble or a data frame with future exogenous variables.

level

The confidence levels (0-100) for the prediction intervals.

quantiles

Quantiles to forecast. Should be between 0 and 1.

finetune_steps

Number of steps used to finetune 'TimeGPT' in the new data.

finetune_loss

Loss function to use for finetuning. Options are: "default", "mae", "mse", "rmse", "mape", and "smape".

clean_ex_first

Clean exogenous signal before making the forecasts using 'TimeGPT'.

add_history

Return fitted values of the model.

model

Model to use, either "timegpt-1" or "timegpt-1-long-horizon". Use "timegpt-1-long-horizon" if you want to forecast more than one seasonal period given the frequency of the data.

num_partitions

A positive integer, "auto", or NULL specifying the number of partitions. When set to "auto", the number of partitions is equal to the number of available cores. When NULL, it defaults to a single partition.

Value

'TimeGPT”s forecast.

Examples

if (FALSE) { # \dontrun{
  nixtlar::nixtla_set_api_key("YOUR_API_KEY")
  df <- nixtlar::electricity
  fcst <- nixtlar::nixtla_client_forecast(df, h=8, id_col="unique_id", level=c(80,95))
} # }