fmvcd-prec.Rmd

---
title: "PREC - filling in missing values in climate data"
author: "IVA"
date: '24.1.17'
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```
### The calculation of all values of the vector by the neural network
```{r compute}
compute_prec_all <- function(net.train, Time){
  return(compute(net.train, Time)$net.result)
}

```
### filling the gaps
```{r filling}
fillin_missing_data <- function(vector_cli, nnet_cli, NORM_TEMP){
  for(i in (1: length(vector_cli))){
    if(is.na(vector_cli[i])){
      vector_cli[i] <- as.integer(nnet_cli[i] * NORM_TEMP)
    }
  }
  return(vector_cli)
}
```
### The construction of neural models and the calculation of missing data
```{r nnet_cli}
library(neuralnet)
NORM_TEMP <- 450.0; NORM_TIME <- 370.0; NUM_TRAIN <- 65; HIDDEN <- 11
NUM_TRAIN <- 265; HIDDEN <- 32

nnet_prec_cli <- function(vector_cli, vec_type = "temp"){
  if( length(vector_cli) < 365) warning("Your year has fewer than 365 days")
  Temp <- vector_cli  / NORM_TEMP
  Time <- (1:length(Temp)) / NORM_TIME
  year_data <- data.frame(Time, Temp)
  if(nrow(year_data) - nrow(na.omit(year_data)) == 0){
    cat("The vector does not contain missing values\n")
    return(vector_cli)
  } else {
    train <- na.omit(year_data)
    train <- train[sample(nrow(train), NUM_TRAIN), ]
    net.train <- neuralnet(Temp~Time, train, hidden = HIDDEN, threshold = 0.01, linear.output=TRUE)
    cat("Error=", net.train$result.matrix[1],"\n")
    nnet_cli <- compute_all(net.train, Time)
    fillin_vector_cli <- fillin_missing_data(vector_cli, nnet_cli, NORM_TEMP)
  }
  return(fillin_vector_cli)
}
```

## model neural network
```{r model_neural_network}
require(nnet)
require(neuralnet)
#filled_path <- "C:/Users/teipolykom/git/fmvcd/krest/"
filled_path <- "/home/larisa/Dropbox/git/fmvcd/krest/"
setwd(filled_path); getwd()

cli_dataset <-read.table("1967.cli", header=FALSE, sep="")
names(cli_dataset) <- c("day", "month", "year", "prec", "temp");
cli_dataset[cli_dataset$prec == -9999, 4] <- NA
print(summary(cli_dataset[, c(4)]))

prec_vector <- cli_dataset[, 4]
# scale inputs: divide by 191 to get 0-1 range
prec_vector_normalize <- prec_vector / max(na.omit(prec_vector))
plot(1: length(prec_vector), prec_vector_normalize)

train <- data.frame(time= (1: length(prec_vector_normalize)), prec=prec_vector_normalize)
train <- data.frame(time= (1: length(prec_vector)), prec=prec_vector/1.0)
train <- na.omit(train)
nnet.fit <- nnet(prec ~ time, data=train, size=21)
neuralnet.fit <- neuralnet(prec ~ time, train, hidden = 11, threshold = 0.1, linear.output=TRUE)
cat("Error =", neuralnet.fit$result.matrix[1],"\n")
neuralnet.compute <- compute(neuralnet.fit, train[,1])$net.result*191

# multiply 191 to restore original scale
nnet.predict <- predict(nnet.fit)*191.0

prec_stat <- c(0L, 0L, 0L, NA, 4L, 0L, 0L, 2L, NA, NA, 5L, NA, NA, NA, 0L, 
NA, NA, 2L, 0L, 6L, 11L, 8L, 0L, NA, 4L, 0L, NA, 0L, 0L, NA, 
0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, NA, 0L, 0L, 0L, 0L, 0L, 0L, 
0L, 0L, 11L, 8L, 0L, 5L, 2L, NA, 2L, NA, 2L, 0L, 0L, 5L, 4L, 
4L, 22L, 11L, 0L, 0L, 2L, 5L, NA, NA, 16L, 11L, 2L, 2L, 0L, 13L, 
2L, 0L, 0L, 0L, 0L, 2L, 2L, 11L, 5L, 0L, 0L, 0L, 0L, 0L, 0L, 
0L, NA, 0L, 0L, NA, 0L, 0L, 13L, 0L, 0L, 0L, 0L, 32L, 0L, 2L, 
NA, 0L, 0L, NA, NA, NA, 16L, 13L, 160L, 0L, 0L, 0L, NA, NA, 0L, 
0L, 0L, 0L, 0L, 0L, 20L, 60L, 20L, NA, 108L, 40L, 89L, 8L, 0L, 
18L, 30L, 37L, 148L, 54L, 0L, 0L, 40L, 0L, 11L, 40L, 45L, NA, 
33L, 70L, 16L, 0L, 67L, 40L, 0L, 9L, 5L, 0L, 0L, 0L, 18L, 0L, 
2L, 0L, 0L, 0L, 47L, 0L, 0L, 0L, 0L, 0L, 0L, 27L, 16L, 0L, 0L, 
0L, 0L, 96L, 0L, 0L, 0L, 0L, 0L, 18L, 11L, 0L, 4L, 0L, 0L, 0L, 
0L, 0L, 0L, NA, 0L, 0L, 0L, 0L, 0L, NA, 80L, 0L, 0L, 0L, 0L, 
0L, 36L, 89L, 4L, 0L, 120L, 145L, 191L, 5L, 0L, 0L, 0L, 0L, 10L, 
4L, 0L, 0L, 0L, NA, 0L, 0L, 0L, 0L, 0L, 0L, 10L, 0L, 0L, 0L, 
0L, 33L, 2L, 4L, 0L, 9L, 0L, 2L, 0L, 0L, 0L, 0L, NA, 0L, 0L, 
65L, 115L, 100L, 21L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 
0L, 0L, 4L, 0L, 20L, 145L, 9L, 26L, 0L, 0L, 0L, 13L, 0L, 0L, 
47L, 5L, 0L, 5L, 0L, 0L, 0L, 0L, 0L, 0L, 2L, 2L, 0L, 0L, 9L, 
0L, 5L, 5L, 2L, 5L, 5L, 8L, 0L, 2L, 4L, 22L, 6L, 18L, 0L, 4L, 
0L, 0L, 9L, 0L, 0L, 0L, 5L, 0L, 0L, 0L, NA, 4L, 5L, 6L, 27L, 
0L, 2L, 0L, 6L, 13L, NA, 4L, 2L, 0L, 0L, 0L, 0L, NA, 0L, 0L, 
0L, 0L, 5L, 6L, 0L, 0L, 2L, 9L, 8L, 4L, 0L, 16L, 0L, 0L, 2L, 
0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 2L, 0L)


# https://heuristically.wordpress.com/2011/11/17/using-neural-network-for-regression/
# https://www.rdocumentation.org/packages/caretEnsemble/versions/2.0.0/topics/caretEnsemble
# https://www.researchgate.net/post/How_can_I_change_standardized_predicted_values_of_neural_network_into_non-standard_values
# http://www.di.fc.ul.pt/~jpn/r/neuralnets/neuralnets.html
# http://www.parallelr.com/r-deep-neural-network-from-scratch/
# https://github.com/PatricZhao/ParallelR
# http://playground.tensorflow.org/#activation=tanh&batchSize=10&dataset=circle&regDataset=reg-plane&learningRate=0.03&regularizationRate=0&noise=0&networkShape=4,2&seed=0.26118&showTestData=false&discretize=false&percTrainData=50&x=true&y=true&xTimesY=false&xSquared=false&ySquared=false&cosX=false&sinX=false&cosY=false&sinY=false&collectStats=false&problem=classification&initZero=false&hideText=false
# http://beyondvalence.blogspot.ru/2014/04/r-comparing-multiple-and-neural-network.html
# http://machinelearningmastery.com/non-linear-regression-in-r/
#http://www.dataminingblog.com/standardization-vs-normalization/
```
```{r}
library(nnet)
# load data
data(longley)
x <- longley[,1:6]
y <- longley[,7]
# fit model
fit <- nnet(Employed~., longley, size=12, maxit=500, linout=T, decay=0.01)
# summarize the fit
summary(fit)
# make predictions
predictions <- predict(fit, x, type="raw")
# summarize accuracy
mse <- mean((y - predictions)^2)
print(mse)
```

## 
  
```{r libs, echo=FALSE}
standardize <- function(){
  # standardize them in range of 0 to 1
  (x-xmin)/(xmax-xmin)
}
# https://jsbin.com
# http://css3generator.com/
# http://enjoycss.com/
# http://www.css3maker.com/
# 

```