gorkang / jspsychhelper Goto Github PK

We need to develop a few things to process Experimental tasks:

• Implement a template where the between and within columns are taken into account
• Make sure to export both wide and long versions of the output.
• save_files(): include parameter to trigger saving both wide and long versions
• Final DF with experimental task in long version and non-experimental wide tasks added?
• condition_between and condition_within SHOULD include task name? If we have more than one experimental task in the protocol, will MISS INFORMATION

Example FONDECYT2022E1

create_targets_file(folder_data, and folder_tasks) can't be used from run_initial_setup

Review and adapt lines_DIRd, function_DIRt, names_dimensions, etc.

If the experiment code is wrong, sometimes all the responses are coded the same.

Implement a CHECK to catch this.

Storing data files in a tar with gzip compression has some important advantages:

• Space reduction and less files (e.g. 9765 files: 35.6 MB -> 3.1 MB)

Drawbacks:

• Time to compress raw files (e.g. 9765 files: 3s) (see below for other compression algorithms)
• Time to decompress and delete with each project run: (e.g. 9765 files: 7.2s now -> 7.7s zip)

See initial implementation: dev/read_ZIP.R

Compression

bzip2: 5.91s | 1.9 Mb
gzip: 3.01s | 3 Mb
xz: 12.05s | 2 Mb
none: 3.37s | 41.3 Mb

In multi-select questions such as DEMOGR_19 Q0 remains in response if response empty.

Careful with create_clean_data changes, as they can easily interfere with separate_responses() resulting in errors in PRFBM_04 and similar items with multiple responses per screen.

The column condition_within should be in the csv outputs when one or more within variables are used.

The within variables names stored in condition_within SHOULD be in camelCase separated by "_", so we can process them.

Examples:

• Good condition_within content: conditionOne_conditionTwo
• Bad condition_within content: conditionOne_condition number two.

When we find Bad names, a potential solution is something like this below, but it is slooooow because first splits each vector for each row, transforms to camelCase and finally pastes the strings together separated by "_":

DFX = DF_clean %>% filter(experimento == "FONDECYT") %>% select(trialid, condition_within) 

DFX %>% 
  rowwise() %>% 
  mutate(condition_within_camel = paste(map_chr(str_split(condition_within, pattern = "_", simplify = TRUE), janitor::make_clean_names, case = "small_camel"), collapse = "_"))

Maybe trigger this ONLY if we find spaces or "-" in the condition_within column?

See new tasks in dev/ALL_PROTOCOLS.R

Updated on 2022-06-14:

• BDI
• CEL
• CGT
• CRQ
• DEMOGR3
• DEMOGRfondecyt2022E1
• ESV
• HC
• ICvsID
• ITC
• LOT
• LSNS
• MCA
• MCQ30
• MDDF
• MLQ
• NARS
• PERMA
• PPD
• ProQOLCV
• PSC
• RobToM
• TCS
• UCLA
• Words
• ZARIT

Buscar manera de crear una sola prueba DEMOGR con TODAS las alternativas habitualmente usadas. Crear sistema para seleccionar subconjunto de preguntas.

Already have part of this in report_DF_clean (Details of tasks; Dimensions).

Include a section in each task correction script with a description of the task and of the dimensions. See prepare_IRI:

• description_task = ""
• description_dimensions = c()

Avoid grepl in # Create long DIR chunk, as it can be ambiguous. Lots of tasks use this now.

Example:

• grepl("01|02", trialid) & RAW == "Varias veces a la semana" ~ 1,

Matches TASK_01, TASK_011, TASK_012...

Will cause issues when we generalize the use of 3 digit trialid's

• All names, dimensions, etc. at begining.
• Separate DF_wide_RAW and DF_wide_RAW_DIR

A copy of the full protocol (jsPsychMaker) should be available zipped.

1. For reproducibility
2. Can check questions with if_question for tests

Maybe the run_initial_setup() could download not only the data but also the full protocol and zip it

See the alpha help page for more details: https://search.r-project.org/CRAN/refmans/psych/html/alpha.html

Probably OK to keep alpha for some uses, but omega should be the default.

The main issue is that the outputs are quite different. Would need someone familiar with the use of omega to set some sane defaults, if possible.

Should add the output of these lines to test-snapshots.R:

  # CREATE SNAPSHOTS TEST LINES
    # FILES_TO_SNAPSHOT = list.files(path = "outputs/data", pattern = "df_.*csv$", full.names = TRUE, ignore.case = FALSE)
    # FILES_TO_SNAPSHOT %>% walk(~ cat(paste0('testthat::expect_snapshot_file(here::here("', .x, '"))\n')))

Right now, the system checks for all the snapshots of canonical protocol!

• template to create R project
• function to build _targets.R reading a maker project folder or an index.html

1. Se decidió dejar el instrumento con 26 items para tener la opción de observar en estos datos cómo se comportan los items. Así que está bien que tenga 26 preguntas el instrumento.
2. El cálculo de dimensiones también está correcto, se calculan en base a la propuesta del paper, a menos que quien analice esta escala en el futuro diga lo contrario y decida usar los items originales, pero por ahora se mantiene así.
3. Lo único que hay que revisar es que la sumatoria del puntaje total sea en base a la suma de las dimensiones y no de los 26 items.

Will need to decompress, move the files, and compress again in data and in .vault/

Maybe just zip it and unzip when testing?

We can run a set of targets using THIS:

targets::tar_make(c('target1', 'target2'), callr_function = NULL)

DO NOT MARK TARGETS as Up to date

targets::tar_make(c('DF_analysis'), callr_function = NULL)

MARK TARGETS as Up to date

targets::tar_make(c('DF_analysis'))

Ahora mismo, si el participante recarga la web, continua donde lo dejo (reinicia la ultima prueba no completada).

Un problema asociado es que, un participante puede reintentar una tarea con F5.

Deberiamos tener un parametro similar a accept_discarded para continuar o no donde lo dejaron?

• Es posible mantener un archivo temporal oculto para cada participante donde se guarda el csv de salida linea a linea ? De ese modo, podriamos continuar exactamente donde lo dejaron...

See ropensci/targets#249 for a description and a reproducible example.

See: https://testthat.r-lib.org/articles/snapshotting.html

Function to create/update/delete snapshots of the state of all df_test and main DF's.

See: https://journals.sagepub.com/doi/10.1177/25152459221128319

The data blinding versions would have the DV column shuffled, and the rest of the dataset untouched, so the analysis can be prepared without knowing about the actual relationships of the IV with the DV.

Think about:

• a way to password protect the original dataset and link the password to the publication of the analysis plan?
• a way to log the access to the original dataset

run_initial_setup() should move those tasks to .vault, etc.

Example, FORM5 in protocol 20

And folder_tasks should work with the zip file of the protocol we create with run_initial_setup

jsPsychHelpeR::create_new_task("Bayesian31")
Error in jsPsychHelpeR::create_new_task():
! R_tasks/prepare_TEMPLATE.R does NOT exist! You can download from
https://raw.githubusercontent.com/gorkang/jsPsychHelpeR/master/R_tasks/prepare_TEMPLATE.R

SHOULD NOT delete the prepare_TEMPLATE() when cleaning up unused tasks from a project...