library(tidymodels)
library(tidyverse)
library(patchwork)
library(ggspatial)
library(modeldata)
library(ggridges)
library(readxl)
library(fco2r)
library(skimr)
library(ISLR)
library(vip)
anomalias <- readr::read_rds("data/anomalias.rds")
glimpse(anomalias)
#> Rows: 79,494
#> Columns: 16
#> $ tratamento <chr> "UC", "UC", "UC", "UC", "UC", "UC", "UC", "UC", "UC", "UC",…
#> $ lat <dbl> -8.620781, -8.620781, -8.620781, -8.869740, -8.869740, -8.8…
#> $ long <dbl> -62.33607, -62.83555, -72.07608, -54.59401, -61.33709, -61.…
#> $ x <dbl> 4082521, 4027557, 3011106, 4934661, 4192994, 4165529, 40831…
#> $ y <dbl> 9036687, 9035452, 8999656, 9019104, 9011189, 9010638, 90088…
#> $ data <dttm> 2015-01-01, 2015-01-01, 2015-01-01, 2015-01-01, 2015-01-01…
#> $ mês <dbl> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,…
#> $ ano <dbl> 2015, 2015, 2015, 2015, 2015, 2015, 2015, 2015, 2015, 2015,…
#> $ estacao <chr> "chuvosa", "chuvosa", "chuvosa", "chuvosa", "chuvosa", "chu…
#> $ xco2 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA,…
#> $ sif <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA,…
#> $ ndvi <dbl> 0.489, 0.664, 0.488, 0.813, 0.636, 0.700, 0.577, 0.891, 0.5…
#> $ lai <dbl> 2.300, 2.975, 2.900, 5.550, 3.250, 1.900, 1.400, 3.350, 2.0…
#> $ lst_amp <dbl> NA, NA, 10.89, 7.38, 5.02, 4.14, NA, NA, NA, NA, 4.46, NA, …
#> $ mediana <dbl> 403.04, 403.04, 403.04, 403.04, 403.04, 403.04, 403.04, 403…
#> $ anomalia <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA,…anomalias %>%
mutate(
fct_ano = fct_rev(as.factor(ano)),
classe = ifelse(tratamento == "UC_desm" | tratamento == "TI_desm",
"Des","Con")
) %>%
ggplot(aes(y=fct_ano)) +
geom_density_ridges(rel_min_height = 0.03,
aes(x=xco2, fill=classe),
alpha = .6, color = "black", from = 395, to = 413
) +
scale_fill_cyclical(values = c("#238B45","#ff8080"),
name = "classe", guide = "legend") +
theme_ridges() 
anomalias %>%
mutate(
fct_ano = fct_rev(as.factor(ano)),
classe = ifelse(tratamento == "UC_desm" | tratamento == "TI_desm",
"Des","Con")
) %>%
ggplot(aes(y=fct_ano)) +
geom_density_ridges(rel_min_height = 0.03,
aes(x=anomalia, fill=classe),
alpha = .6, color = "black", from = -10, to = 10
) +
scale_fill_cyclical(values = c("#238B45","#ff8080"),
name = "classe", guide = "legend") +
theme_ridges() 
lendo a Base para o Aprendizado de Máquina
data_set_ml <- anomalias %>%
mutate(
fct_ano = fct_rev(as.factor(ano)),
classe = as_factor(ifelse(tratamento == "UC_desm" | tratamento == "TI_desm",
"Des","Con")
)) %>% #<-------
drop_na()
classe_initial_split <- initial_split(data_set_ml, prop = 0.75,
strata = classe)classe_train <- training(classe_initial_split)
# fco2_test <- testing(fco2_initial_split)
# visdat::vis_miss(fco2_test)
classe_train %>%
ggplot(aes(x=xco2, y=..density..))+
geom_histogram(bins = 30, color="black", fill="lightgray")+
geom_density(alpha=.05,fill="red")+
theme_bw() +
labs(x="xco2 - treino", y = "Densidade")
classe_train %>%
ggplot(aes(x=sif, y=..density..))+
geom_histogram(bins = 30, color="black", fill="lightgray")+
geom_density(alpha=.05,fill="red")+
theme_bw() +
labs(x="sif - treino", y = "Densidade")
classe_train %>%
ggplot(aes(x=ndvi, y=..density..))+
geom_histogram(bins = 30, color="black", fill="lightgray")+
geom_density(alpha=.05,fill="red")+
theme_bw() +
labs(x="ndvi - treino", y = "Densidade")
classe_train %>%
ggplot(aes(x=lai, y=..density..))+
geom_histogram(bins = 30, color="black", fill="lightgray")+
geom_density(alpha=.05,fill="red")+
theme_bw() +
labs(x="lai - treino", y = "Densidade")
classe_train %>%
ggplot(aes(x=lst_amp, y=..density..))+
geom_histogram(bins = 30, color="black", fill="lightgray")+
geom_density(alpha=.05,fill="red")+
theme_bw() +
labs(x="lst_amp - treino", y = "Densidade")
classe_recipe <- recipe(classe ~ .,
data = classe_train %>%
select(classe, xco2:lst_amp)
) %>%
# step_normalize(all_numeric_predictors()) %>%
# step_naomit() %>%
step_novel(all_nominal_predictors()) %>%
step_zv(all_predictors()) %>%
# step_naomit(c(ts, us)) %>%
#step_impute_median(where(is.numeric)) %>% # inputação da mediana nos numéricos
# step_poly(c(Us,Ts), degree = 2) %>%
step_dummy(all_nominal_predictors())
bake(prep(classe_recipe), new_data = NULL)
#> # A tibble: 3,196 × 6
#> xco2 sif ndvi lai lst_amp classe
#> <dbl> <dbl> <dbl> <dbl> <dbl> <fct>
#> 1 403. 0.8 0.507 1.85 5.83 Con
#> 2 398. 0.912 0.866 3.4 5.78 Con
#> 3 401. 1.15 0.512 1.12 0.92 Con
#> 4 407. 0.393 0.722 2.22 3.26 Con
#> 5 402. 1.25 0.764 3.9 3.94 Con
#> 6 399. 1.16 0.46 1.62 2.52 Con
#> 7 403. 0.891 0.748 2.88 4.36 Con
#> 8 402. 1.07 0.547 3.4 7.84 Con
#> 9 404. 1.49 0.832 4.23 6.71 Con
#> 10 404. 0.81 0.797 4.47 4.61 Con
#> # ℹ 3,186 more rowsvisdat::vis_miss(bake(prep(classe_recipe), new_data = NULL))
classe_nn_model <- mlp(
hidden_units = 2) %>% # margin sempre para regressão
set_mode("classification") %>%
set_engine("nnet") %>%
fit(classe ~ ., data = classe_train %>%
select(classe, xco2:lst_amp))
NeuralNetTools::plotnet(classe_nn_model$fit)
classe_resamples <- vfold_cv(classe_train, v = 5)
classe_nn_model <- mlp(
hidden_units = tune(),
penalty = tune(),
epochs = tune()
) %>% # margin sempre para regressão
set_mode("classification") %>%
set_engine("nnet") classe_nn_wf <- workflow() %>%
add_model(classe_nn_model) %>%
add_recipe(classe_recipe)
grid_nn <- expand.grid(
hidden_units = c(1,2),
penalty = c(1,5),
epochs = c(50, 100)
)
glimpse(grid_nn)
#> Rows: 8
#> Columns: 3
#> $ hidden_units <dbl> 1, 2, 1, 2, 1, 2, 1, 2
#> $ penalty <dbl> 1, 1, 5, 5, 1, 1, 5, 5
#> $ epochs <dbl> 50, 50, 50, 50, 100, 100, 100, 100
classe_nn_tune_grid <- tune_grid(
classe_nn_wf,
resamples = classe_resamples,
grid = grid_nn,
metrics = metric_set(roc_auc)
)
area_nn <- collect_metrics(classe_nn_tune_grid) %>%
filter(.metric == "roc_auc") %>%
summarise(area = mean(mean),
desvio_pad = mean(std_err))autoplot(classe_nn_tune_grid)
classe_nn_best_params <- select_best(classe_nn_tune_grid, "roc_auc")
classe_nn_wf <- classe_nn_wf %>% finalize_workflow(classe_nn_best_params)
classe_nn_last_fit <- last_fit(
classe_nn_wf,
classe_initial_split
)
# Variáveis importantes
classe_nn_last_fit_model <-classe_nn_last_fit$.workflow[[1]]$fit$fit
vip(classe_nn_last_fit_model,
aesthetics = list(color = "black", fill = "orange")) +
theme(axis.text.y=element_text(size=rel(1.5)),
axis.text.x=element_text(size=rel(1.5)),
axis.title.x=element_text(size=rel(1.5))
) +
theme_bw()
classe_test_preds_nn <- collect_predictions(classe_nn_last_fit)
classe_roc_curve_nn <- classe_test_preds_nn %>%
roc_curve(classe, .pred_Con)
autoplot(classe_roc_curve_nn)
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent