RvalPractice3.Rmd

---
title: "Automated report example"
author: "Author"
date: "March 2021"
output:
  pdf_document: default
  word_document: default
  html_document: default
params:
  year:
    label: Year
    value: 2017
    input: slider
    min: 2010
    max: 2021
    step: 1
    sep: ''
  data:
    label: 'Input dataset:'
    value: some_data.csv
    input: file
---

\vspace{0.3cm}

```{r}

# the parameters in the Rmd-file header can be chosen interactively, 
# by using "Knit with parameters"

# or directly, by running:

# rmarkdown::render("RvalPractice0.Rmd", params = list(
#   year = 2021,
#   file = "our_favourite.csv"
# ))

# Try this!

```

# Useful links:

- for downloading and installing R, RStudio on MacOSX, see the example at:
https://web.stanford.edu/~kjytay/courses/stats32-aut2018/Session%201/Installation%20for%20Mac.html 

- for using open data from the web:
https://theodi.org/article/how-to-use-r-to-access-data-on-the-web/

(_hint_: use functions like read.csv(), read.url(), various API's or the RCurl-package )



# Important notes:

- We may all raise issues in the github repository 

https://github.com/violetacln/learnRval  

about any questions, proposals, ideas we like to share!


- We do as much as possible here, in our interactive meeting: 

  - we run the examples included in these RvalPractice - files

  - we could also propose new ones and/or run code on different data.



# Introduction: 

## Data-set and main information needed for validation

about: time-variables, modeled - variables, imputed-variables, if the case

Assume a data-frame has been built.

Assume rules set has been built.

Example of notations: 
  
  ```{r datadef, echo=TRUE, eval=FALSE}
  
## data rules, built, read into a data frame and then a validator object
# vrules <- 

## modifier rules, built, read into a data frame and then a validator object
# mrules <- 
# v <-

## main data set as a data.frame
#df <- 
# data <- 

##data sets which need to be compared (could be same data measured at 
##different moments in time), as data.frames
#df1<- 
#df2<-

##time series we need to check, if it is ths case
#tsuniv
#tsmultiv
```

__Important note:  when we want to create our report and show all results, we make "eval=TRUE" in the R-chunks which follow.__





# Part 3: Take away exercises on: error correction, complex output validation, Bayesian validation 




## Error Detection: more on error location 

Which field values should we correct, on each record?


```{r error location, echo=TRUE, eval=FALSE}

library(errorlocate)

rules <- validator(price >500, carat > 0.2)   # check summary!
data <- ggplot2::diamonds

  error_locations <- locate_errors(data, rules)
  head(values(error_locations))  
  summary(error_locations)
 
data_marked_errors <- replace_errors(data, rules)

# faulty data was replaced with NA
print(data_marked_errors)
er <- errors_removed(data_marked_errors)
#print(er)
summary(er)
#er$errors
sum(is.na(data))
sum(is.na(data_marked_errors))  # our values to be imputed!



```




## Error correction: apply modifier rules

```{r modifier rules, echo=TRUE, eval=FALSE}

library(dcmodify)

df <- ggplot2::diamonds
m <- modifier(if (price <= 500) price <- 500)

modified <- modify(df, m)
head(modified, 3)


```



## Error correction: modifier rules discovery option: association rules discovery
```{r association rules, echo=TRUE, eval=FALSE}

df  <- ggplot2::diamonds[,2:4]
#(citation("arules"))
library(arules)   
tdata <- as(df, "transactions")

#method 1 of clustering: eclat algorithm
eclat_res <- inspect(eclat(tdata, 
                           parameter = list(supp=0.07, maxlen=15)))
eclat_plot <- itemFrequencyPlot(tdata, topN=10, 
                          type="absolute", main="item freguency")

summary_data <- summary(tdata)

eclat_summary <- summary(eclat(tdata, 
                          parameter = list(supp=0.07, maxlen=15)))


#method 2 of clustering: apriori algorithm
rules <- apriori(tdata)
apriori_summary <- summary(rules)
apriori_res <- inspect(rules)

```



## Error correction: imputation and adjustments


```{r , echo=TRUE, eval=FALSE}

library(simputation)

dat <- ggplot2::diamonds

dat[1:3,1] <- dat[3:7,2]  <- NA
head(dat,10)

da1 <- impute_lm(dat, carat ~ clarity+depth+price)
head(da1,3)

da2 <- impute_median(da1, cut ~ x+y+z)
head(da2,3)

#chaining all these methods is possible, with %>% of magrittr - package

```


## Comparing data sets

```{r , echo=TRUE, eval=FALSE}

vrules <- validate::validator(price >500, carat > 0.2)   # check summary!
input <- ggplot2::diamonds

m <- dcmodify::modifier(if (price <= 500) price <- 500)
cleaned <- dcmodify::modify(df, m)


comparison <- validate::compare(vrules
                    , input , cleaned 
                    )

comparison

par(mfrow=c(2,1))
barplot(comparison)
plot(comparison)

```

Create more comparisons!



# Validation of results with data mining methods


## Identify clusters


```{r clusters, echo=TRUE, eval=FALSE}

df <- diamonds[1:2000,c("carat", "depth", "price")]

# find optimal number of clusters
  dff <- scale(df)
 
 ### factoextra::fviz_nbclust(dff, kmeans, method = "gap_stat")

# compute and visualise
  set.seed(123)
  km.res <- kmeans(dff, 5, nstart = 25)
  # visualize
  factoextra::fviz_cluster(km.res, data = dff,
               ellipse.type = "convex",
               palette = "jco",
               repel = TRUE,
               ggtheme = ggplot2::theme_minimal())

  # try to find an optimum!
  

# compare also with PAM clustering
  # Compute PAM
  pam.res <- cluster::pam(dff, 5)
  # Visualize
  factoextra::fviz_cluster(pam.res)  
  
   # try to find an optimum!
  
```



## Exploring and reviewing models

See the example in the slides of Part 3. Try new data!

```{r models, glm or ts: residuals and goodness of fit, echo=TRUE, eval=FALSE}


# if a glm type of model, then
#rev_model(model_a)
##if a time series model, then
##rev_model_ts(model_a)
```


## Exploring and reviewing time series characteristics

See the example in the slides of Part 3. Try new data!

```{r time series characteristics: univariate and multivariate, echo=TRUE, eval=FALSE}


#for all time series, univariate
#rev_ts_univ(tsuniv)
##for any multivariate time series
##rev_ts_multiv(tsmultiv)
```




# Alternative: Bayesian validation, simultaneous detection and correction of errors

Try this with different parameters and number of iterations. 

Try new data!


```{r Bayesian validation, echo=TRUE, eval=FALSE}

library(EditImputeCont)

## read the toy example data, which has two ratio edits and a balance edit
data(SimpleEx)

data1 = readData(Y.original=SimpleEx$D.obs, ratio=SimpleEx$Ratio.edit, 
	range=NULL, balance=SimpleEx$Balance.edit)

## create and initialize the model with 15 DP mixture components
 model1 = createModel(data.obj=data1, K=15)
 
## Run an iteration of MCMC
model1$Iterate()
dim(model1$Y.edited)
## [1]   1000   4  
# Edit-imputed datasets of n=1000 records with p=4 variables

## Please see the example in the demo folder for more details
```