PASS Summit Notes for my AzureML, R and Power BI Presentation

I’m going to have fun with my AzureML session today at PASS Summit! More will follow on this post later; I am racing off to the keynote so I don’t have long 🙂

I heard some folks weren’t sure whether to attend my session or Chris Webb’s session. I’m honestly flattered but I’m not in the same league as Chris! I’ve posted my notes here so that folks can go off and attend Chris’ session, if they are stuck between the two.

Here is the order of things:

  • Slide Deck
  • How do you choose a machine learning algorithm?
  • How do you carry out an AzureML project?
  • AzureML Experiment 
  • R Code

So, the slide deck is here:

  • AzureML Experiment 

You can see this experiment in the AzureML Gallery. You may have to sign up for a Windows Live account to get a free AzureML studio account, and I recommend that you do.

  • How do you choose a machine learning algorithm?

Kudos to Microsoft – this is their cheatsheet and I recommend that you look at the original page.

Here is some more information on the topic from Microsoft, and I recommend that you follow it.

How do you carry out an AzureML project?

Try the CRISP-DM Framework for a start

See the Modelling Agency for the original source. https://the-modeling-agency.com/crisp-dm.pdf

CRISP-DM Process Diagram.png
CRISP-DM Process Diagram” by Kenneth JensenOwn work. Licensed under CC BY-SA 3.0 via Commons.

R Code

Here’s a sample R code. I know it is simple, and there are better ways of doing this. However, remember that this is for instructional purposes in front of +/- 500 people so I want to be sure everyone has a grounding before we talk more complicated things.

You may have to install the libraries first, if you haven’t done so.

library(data.table)
library(ggplot2)
library(xtable)
library(rpart)
require(xtable)
require(data.table)
require(ggplot2)
require(rpart)

summary(adult.data)
class(adult.data)

# Let’s rename the columns
names(adult.data)[1]<-“age”
names(adult.data)[2]<-“workclass”
names(adult.data)[3]<-“fnlwgt”
names(adult.data)[4]<-“education”
names(adult.data)[5]<-“education.num”
names(adult.data)[6]<-“marital.status”
names(adult.data)[7]<-“occupation”
names(adult.data)[8]<-“relationship”
names(adult.data)[9]<-“race”
names(adult.data)[10]<-“sex”
names(adult.data)[11]<-“capital.gain”
names(adult.data)[12]<-“capital.loss”
names(adult.data)[13]<-“hours.per.week”
names(adult.data)[14]<-“country”
names(adult.data)[15]<-“earning_level”

# Let’s see if the columns renamed well
# What is the maximum age of the adult?
# How much data is missing?
summary(adult.data)

# How many rows do we have?
# 32561 rows, 15 columns
dim(adult.data)

# There are lots of different ways to deal with missing data
# That would be a session in itself!
# For demo purposes, we are simply going to replace question marks, and remove rows which have anything missing.

adult.data$workclass <- as.factor(gsub(“[?]”, NA, adult.data$workclass))
adult.data$education <- as.factor(gsub(“[?]”, NA, adult.data$education))
adult.data$marital.status <- as.factor(gsub(“[?]”, NA, adult.data$marital.status))
adult.data$occupation <- as.factor(gsub(“[?]”, NA, adult.data$occupation))
adult.data$relationship <- as.factor(gsub(“[?]”, NA, adult.data$relationship))
adult.data$race <- as.factor(gsub(“[?]”, NA, adult.data$race))
adult.data$sex <- as.factor(gsub(“[?]”, NA, adult.data$sex, fixed = TRUE))
adult.data$country <- as.factor(gsub(“[?]”, NA, adult.data$country))

is.na(adult.data) = adult.data==’?’
is.na(adult.data) = adult.data==’ ?’
adult.tidydata = na.omit(adult.data)

# Let’s check out our new data set, called adult.tidydata
summary(adult.tidydata)

# How many rows do we have?
# 32561 rows, 15 columns
dim(adult.tidydata)

# Let’s visualise the data
boxplot(adult.tidydata$education.num~adult.tidydata$earning_level,outline=F,xlab=”Income Level”,ylab=”Education Level”,main=”Income Vs Education”)

prop.table(table(adult.tidydata$earning_level,adult.tidydata$occupation),2)
for (i in 1:ncol(adult.tidydata)-2) {
if (is.factor(adult.tidydata[,i])){
pl =ggplot(adult.tidydata,aes_string(colnames(adult.tidydata)[i],fill=”earning_level”))+geom_bar(position=”dodge”) + theme(axis.text.x=element_text(angle=75))
print(pl)
}

}

evalq({
plot <- ggplot(data = adult.tidydata, aes(x = hours.per.week, y = education.num,
colour = hours.per.week))
plot <- plot + geom_point(alpha = 1/10)
plot <- plot + ggtitle(“Hours per Week vs Level of Education”)
plot <- plot + stat_smooth(method = “lm”, se = FALSE, colour = “red”, size = 1)
plot <- plot + xlab(“Education Level”) + ylab(“Hours per Week worked”)
plot <- plot + theme(legend.position = “none”)
plot
})

That’s all for now! More later.

Jen xx

One thought on “PASS Summit Notes for my AzureML, R and Power BI Presentation

  1. Pingback: News About PASS For Week Ending October 31, 2015 |

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s