robyn_refresh assumes default model path

[rocker-david]

Project Robyn

Describe issue

When I save trained model to a non-default path, I get the following behaviors:

  robyn_write(
    InputCollect,
    OutputCollect,
    select_model = sol_id,
    dir = paste(OutputCollect$plot_folder,"models",sep = "")
  )

Using a custom path (e.g. outputs/Robyn_202303101230_init/models/RobynModel-2_16_5.json) I can load it back no problem.

pretrained_model_json_file <- "outputs/Robyn_202303101230_init/models/RobynModel-2_16_5.json"
json_obj <- robyn_read(pretrained_model_json_file)
>>> Recreating model 2_16_5
Imported JSON file succesfully: outputs/Robyn_202303101230_init/models/RobynModel-2_16_5.json

then I run robyn_refresh

RobynRefresh <- robyn_refresh(
  json_file = pretrained_model_json_file,
  dt_input = refresh_df,
  dt_holidays = custom_holidays,
  refresh_steps = 30,
  refresh_iters = 200,
  refresh_trials = 5,
  ts_validation = TRUE,
  plot_folder = robyn_output_dir
)
>> Running feature engineering...
Input data has 730 days in total: 2019-01-01 to 2020-12-30
Initial model is built on rolling window of 700 day: 2019-01-01 to 2020-11-30
Successfully recreated model ID: 2_16_5
Error in difftime(totalDates, refreshStart, units = "days") : 
  object 'refreshStart' not found

I'm able to get around this by adding

refreshStart <- '2019-01-01'
refreshEnd <- '2020-12-30'

now it successfully runs and saves outputs
but breaks before returning the RobynRefresh object successfullly with the following message

...
>> Exported model inputs as ./outputs/Robyn_202303101230_init/Robyn_202303101355_rf1/RobynModel-inputs.json
- DECOMP.RSSD NOT converged: sd@qt.20 0.13 > 0.036 & |med@qt.20| 0.037 > 0.014
- NRMSE NOT converged: sd@qt.20 0.21 > 0.063 & |med@qt.20| 1.4 > 1.4
Total run time: 1.3 mins
Selected model ID: 1_10_8 for refresh model #1 based on the smallest combined normalised errors
Error in robyn_read(filename, quiet = TRUE) : 
  JSON file can't be imported: ./outputs/Robyn_202303101230_init/RobynModel-2_16_5.json

I noticed I'm saving to the regular output path, with an added models folder nested, but this exception assumes the filename to be in the default path:

• outputs/Robyn_202303101230_init/models/RobynModel-2_16_5.json
• outputs/Robyn_202303101230_init/RobynModel-2_16_5.json

I think the error may be happening as it assumes the filename and passes to robyn_read

• Robyn/R/R/json.R

  Line 287 in f613902

  filename <- paste(c(base_dir, chain[1:i], file), collapse = "/")

I've been able to get around and successfully run the function by saving the model file to the assumed path.
I don't know if this is an actual bug or if we're not supposed to save / load the models anywhere else other than the default path but was hoping to see if we can just pass the json_file path directly to robyn_read

Provide reproducible example

I'm using dummy daily sample data I can provide if requested, but I think this should be reproduce-able with the code provided above as I reproduced it with both sample data and real data.

Environment & Robyn version

Make sure you're using the latest Robyn version before you post an issue.

• Check and share Robyn version: packageVersion("Robyn")
  • 3.9.0
• R version (Please, check and share: sessionInfo() or R.version$version.string)
  • 4.2.1

Thanks in advance and big kudos/thanks to the Robyn team!

[rocker-david]

having the model saved to the assumed path (without the models subdir) works