# DAL ToolBox
# version 1.0.777

source("https://raw.githubusercontent.com/cefet-rj-dal/daltoolbox/main/jupyter.R")

#loading DAL Toolbox
load_library("daltoolbox")

#load required library
library(ggplot2)

i <- seq(0, 25, 0.25)
x <- cos(i)

plot_ts(x=i, y=x) + theme(text = element_text(size=16))

sw_size <- 10
ts <- ts_data(x, sw_size)
ts_head(ts, 3)

test_size <- 1
samp <- ts_sample(ts, test_size)
ts_head(samp$train, 3)
ts_head(samp$test)

# Setup for tunning using ELM
tune <- ts_tune(input_size=c(3:5), base_model = ts_elm(ts_norm_gminmax()))
ranges <- list(nhid = 1:5, actfun=c('sig', 'radbas', 'tribas', 'relu', 'purelin'))

# tune <- ts_tune(input_size=c(3:5), base_model = ts_lstm(ts_norm_gminmax()))
# ranges <- list(input_size = 1:10, epochs=10000)

io_train <- ts_projection(samp$train)

# Generic model tunning
model <- fit(tune, x=io_train$input, y=io_train$output, ranges)

adjust <- predict(model, io_train$input)
ev_adjust <- evaluate(model, io_train$output, adjust)
print(head(ev_adjust$metrics))

steps_ahead <- 1
io_test <- ts_projection(samp$test)
prediction <- predict(model, x=io_test$input, steps_ahead=steps_ahead)
prediction <- as.vector(prediction)

output <- as.vector(io_test$output)
if (steps_ahead > 1)
    output <- output[1:steps_ahead]

print(sprintf("%.2f, %.2f", output, prediction))

ev_test <- evaluate(model, output, prediction)
print(head(ev_test$metrics))
print(sprintf("smape: %.2f", 100*ev_test$metrics$smape))

yvalues <- c(io_train$output, io_test$output)
plot_ts_pred(y=yvalues, yadj=adjust, ypre=prediction) + theme(text = element_text(size=16))

### Ranges for ELM
ranges_elm <- list(nhid = 1:20, actfun=c('sig', 'radbas', 'tribas', 'relu', 'purelin'))

### Ranges for MLP
ranges_mlp <- list(size = 1:10, decay = seq(0, 1, 1/9), maxit=10000)

### Ranges for RF
ranges_rf <- list(nodesize=1:10, ntree=1:10)

### Ranges for SVM
ranges_svm <- list(kernel=c("radial", "poly", "linear", "sigmoid"), epsilon=seq(0, 1, 0.1), cost=seq(20, 100, 20))

### Ranges for LSTM
ranges_lstm <- list(input_size = 1:10, epochs=10000)

### Ranges for CNN
ranges_cnn <- list(input_size = 1:10, epochs=10000)