ctry-specific model#5
Conversation
jonsampedro
left a comment
jonsampedro
left a comment
There was a problem hiding this comment.
Casi todo son comentarios menores, pero el tema del HIA_Adder es incrrecto y hay que quitralo... comentamos!
| @@ -1,49 +1,74 @@ | |||
| name: build | |||
There was a problem hiding this comment.
Algo le pasa a esta acción que no la pasa (Ubuntu)
| @@ -1,26 +1,21 @@ | |||
| name: test_coverage | |||
There was a problem hiding this comment.
Lo mismo, da algún error esta acción
| years_in_prj <- listYears(prj) | ||
| years_in_prj <- years_in_prj[!is.na(years_in_prj)] | ||
| years_in_prj <- setdiff(years_in_prj, NA) | ||
| base_year <- dplyr::if_else(2021 %in% years_in_prj, 2021, 2015) |
There was a problem hiding this comment.
Con esto se supone que funciona para GCAM (>7.0) que tenga o BY 2021 o 2015, no? Igual poner una nota?
| # Get model to subtract coefficients and predict | ||
| fit_model_output_c <- fit_model(HIA_var = HIA_var, countries = c) | ||
| # Safe check for non-complete country datasets | ||
| if (is.null(fit_model_output_c) || length(fit_model_output_c) == 0) next |
There was a problem hiding this comment.
Se salta el país que no tiene datos para hacer el modelo no? Por curiosidad, podemos ver si esto pasa en muchos sitios con el actual panel data?
There was a problem hiding this comment.
pasa en
[1] "Bahamas"
[1] "Belize"
[1] "Bhutan"
[1] "Brunei Darussalam"
[1] "Eritrea"
[1] "Fiji"
[1] "Grenada"
[1] "Guyana"
[1] "Kiribati"
[1] "Maldives"
[1] "Papua New Guinea"
[1] "Solomon Islands"
[1] "Somalia"
[1] "Sudan"
[1] "Suriname"
[1] "Tonga"
[1] "Vanuatu"
les faltan datos de gdp
| output_fin_c <- output_c %>% | ||
| dplyr::mutate(pred_var = paste0("pred_log_", HIA_var, "_per_100K")) %>% | ||
| gcamdata::left_join_error_no_match( | ||
| rhap::hia_adder %>% |
There was a problem hiding this comment.
Esto es lo que creo que hay que cambiar: el hia_ader era un ajuste que se hacía con el modelo global (que ya no usamos):
bias.adder = deaths observadas en 2020 - estimadas en 2020 con el modelo global
Luego se lo sumabamos aquí por mejorar la estimación, pero al usar el fit_model por cada país esto solo va a distrsionar los resultados. creo que hay que uquitar esta parte, borrar este input de hia_adder del data del paquete y ver si queremos usar un bias.adder en nuestras proyecciones, pero en todo caso un adder calculado en cada pais con el modelo del pais!
| pred_var = gsub("log_", "", pred_var) | ||
| ) %>% | ||
| dplyr::mutate( | ||
| pred_value_per_100K_adj = pred_value_per_100K + bias.adder, |
| # Restrict data to selected countries | ||
| if (countries != 'All') output_gr <- output_gr %>% dplyr::filter(country_name %in% countries) | ||
|
|
||
| output.panel.gr <- plm::pdata.frame(output_gr, index = c("country_name", "year")) |
There was a problem hiding this comment.
No puedo ver si en el "by group" stá el hia adder pero no sería correcto tampoco aquí
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #5 +/- ##
==========================================
- Coverage 87.36% 87.32% -0.04%
==========================================
Files 7 8 +1
Lines 1171 1341 +170
==========================================
+ Hits 1023 1171 +148
- Misses 148 170 +22 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
2 participants
Objective: FE model improvement considering each model separately. The idea is to loop across all countries (indicated by the user, "All" by default) to fit and use for prediction ctry-specific models. The predicted vs actual values plot is awasome. The RMSE values are around 0.1 everywhere
FE_all_ctries_new.pdf