Skip to content

Why does my saved calibration data load as environment variable and not, when I save and reload from rds? #51

Open
@leonhGeis

Description

@leonhGeis

In order to save my gluonts calibration tbl, I want to be able to reload the saved calibration table from an rds temp file.

As I know, there are convenience functions to save gluonts models, but I rather want to save the calibrated data. This is mainly because, when I refit the model, gluonts is changing the forecast values, as they are probabilistic, every time. To get a model history, I want to save calibration data and the refit table from the ex ante.

I tried to do this, but every time, I want to reload the saved calibration from rds, the forecasting function fails with an error.

I supplied a reprex to illustrate the issue:

library(tidymodels)
library(tidyverse)
library(timetk)
library(reticulate
)
Sys.setenv(GLUONTS_PYTHON = "/YOUR_PATH.../r-miniconda/envs/r-reticulate/bin/python3.8"
)
library(modeltime.gluonts
)
HORIZON <- 24*7
data <- m4_hourly %>%
	select(id, date, value
		) %>%
	group_by(id
		)%>%
	mutate(value = standardize_vec(value)
		) %>%
	ungroup()
new_data <- data %>%
	group_by(id
		) %>%
	future_frame(.length_out = HORIZON
		) %>%
	ungroup()
model_fit_nbeats_ensemble <- nbeats(
	id                    = "id",
	freq                  = "H",
	prediction_length     = HORIZON,
	lookback_length       = c(HORIZON, 4*HORIZON),
	epochs                = 5,
	num_batches_per_epoch = 15,
	batch_size            = 1 
	) %>%
	set_engine("gluonts_nbeats_ensemble"
		) %>%
	fit(value ~ date + id, data)
set.seed(223
)
calibration_tbl <- 
	modeltime_table(
		model_fit_nbeats_ensemble
		) %>%
	modeltime_calibrate(
		new_data = data, id = "id"
		) 
# save calibrated model
artifacts <- list(
	calibration = calibration_tbl   # ---> USING THIS AS FC-INPUT WORKS
)
artifacts %>%
	write_rds("artifacts_list.rds"
	)
artifacts <- read_rds("artifacts_list.rds" # ---> GET SAVED MODEL FAILS
)
# forecast test
set.seed(273
)
forecast <- artifacts$calibration  %>% 
	modeltime_forecast(
		new_data      = new_data,
		actual_data   = data,
		keep_data     = TRUE
	)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions