Assessment_via_cluster: Calculate error metrics for all algorithm per cluster

In bishun945/FCMm: Fuzzy Cluster Method Based on the Optimized m Value (Fuzzifier)

Description

Calculate error metrics for all algorithm per cluster

Usage

Assessment_via_cluster(
  pred,
  meas,
  memb,
  metrics = c("MAE", "MAPE"),
  log10 = FALSE,
  total = TRUE,
  hard.mode = TRUE,
  cal.precision = FALSE,
  valid.definition = list(negative = FALSE, percent = 0.6),
  na.process = FALSE,
  plot.col = FALSE
)

Arguments

pred	prediction of Chla
meas	in-situ measurement of Chla
memb	membership value matrix
metrics	metrics need to be calculated
log10	Should pred and meas be log10-transformed? (default as FALSE)
total	Whether to calculate summarized metrics (default as TRUE)
hard.mode	If FALSE, the membership values are used to calculate validation metrics
cal.precision	Whether to calculate precision (only support for vectorized metrics), default as FALSE
valid.definition	The definition of valid prediction, default as list(negative=FALSE, percent = 0.6). The invalid prediction will not be removed to calculate error metrics. negative=FALSE means the negative values are considered as invalid predictions while percent means the tolerance of error-percentage that only the predictions between (y-y*percent), y+y*percent) are valid.
na.process	na.process and choose to statistic NA value percent
plot.col	option to plot col result for selected metrics (default as FALSE)

Value

Results of Assessment_via_cluster are returned as a list including:

Metrcs	A list of the selected metric values for all algorithms. Valid_percent would be included if na.process are set as TRUE
res_plot	Bar plots by using ggplot function for metrics value at every cluster.
res_plot_dt	Dataframe for plotting res_plot. I just keep it in case of plotting other types
res_plot_facet	res_plot added on facet_wrap.
input	input parameters of Assessment_via_cluster

Note

If the cal.precision is TRUE, the hard.mode == TRUE is used. In that case, mean and sd calculation is conducted for hard mode based on result from cal.metrics.vector.

See Also

Other Algorithm assessment: Getting_Asses_results(), Sampling_via_cluster(), Score_algorithms_interval(), Score_algorithms_sort(), Scoring_system()

Examples

library(FCMm) 
library(ggplot2) 
library(magrittr)
library(stringr)
data("Nechad2015")
x <- Nechad2015[,3:11]
wv <- gsub("X","",names(x)) %>% as.numeric
set.seed(1234)
w <- sample(1:nrow(x), 100) 
x <- x[w, ]
names(x) <- wv
nb = 4 # Obtained from the vignette "Cluster a new dataset by FCMm"
set.seed(1234)
FD <- FuzzifierDetermination(x, wv, stand=FALSE)
result <- FCM.new(FD, nb, fast.mode = TRUE)
p.spec <- plot_spec(result, show.stand=TRUE)
print(p.spec$p.cluster.spec)
Chla <- Nechad2015$X.Chl_a..ug.L.[w]
Chla[Chla >= 999] <- NA
dt_Chla <- run_all_Chla_algorithms(x) %>% as.data.frame
dt_Chla <- data.frame(Chla_true = Chla, 
BR_Gil10 = dt_Chla$BR_Gil10, 
OC4_OLCI = dt_Chla$OC4_OLCI, 
OCI_Hu12 = dt_Chla$OCI_Hu12, 
NDCI_Mi12= dt_Chla$NDCI_Mi12) %>% round(3)
w = which(!is.na(dt_Chla$Chla_true))
dt_Chla = dt_Chla[w,]
memb = result$res.FCM$u[w,] %>% round(4)
Asses_soft <- Assessment_via_cluster(pred = dt_Chla[,-1], 
meas = dt_Chla[,1], memb = memb, log10 = TRUE, hard.mode = FALSE, 
na.process = TRUE, plot.col = TRUE)
Asses_soft$res_plot_facet
knitr::kable(Asses_soft$MAE %>% round(3))
knitr::kable(Asses_soft$MAPE %>% round(2))

bishun945/FCMm documentation built on Oct. 15, 2021, 6:43 p.m.