SLIDE 1
DataCamp Supervised Learning in R: Case Studies
Surveying Catholic sisters in 1967
SUPERVISED LEARNING IN R: CASE STUDIES
DataCamp Supervised Learning in R: Case Studies
Surveying Catholic sisters in 1967 Julia Silge Data Scientist at - - PowerPoint PPT Presentation
Surveying Catholic sisters in 1967 Julia Silge Data Scientist at - - PowerPoint PPT Presentation
DataCamp Supervised Learning in R: Case Studies SUPERVISED LEARNING IN R : CASE STUDIES Surveying Catholic sisters in 1967 Julia Silge Data Scientist at Stack Overflow DataCamp Supervised Learning in R: Case Studies Conference of Major
SLIDE 2
SLIDE 3
DataCamp Supervised Learning in R: Case Studies
Opinions and attitudes in the 1960s
Response Code Disagree very much 1 Disagree somewhat 2 Neither agree nor disagree 3 Agree somewhat 4 Agree very much 5
Check out the survey's . codebook
SLIDE 4
DataCamp Supervised Learning in R: Case Studies
Opinions and attitudes in the 1960s
> sisters67 # A tibble: 77,112 x 67 age sister v116 v117 v118 v119 v120 v121 v122 v123 v124 v125 <dbl> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int> 1 60.0 1 1 1 3 5 1 1 3 5 3 1 2 70.0 2 2 2 4 4 1 3 1 5 4 1 3 60.0 3 1 1 3 2 2 3 1 1 3 1 4 60.0 4 5 1 2 4 1 3 4 3 3 4 5 50.0 5 2 3 3 3 2 2 1 5 2 5 6 40.0 7 4 3 2 5 4 3 1 5 2 5 7 50.0 9 5 4 5 4 4 5 3 5 4 2 8 40.0 10 5 4 3 5 1 3 5 5 5 4 9 30.0 11 2 2 3 5 1 3 3 5 1 3 10 30.0 12 4 1 5 5 1 4 3 5 1 5 # ... with 77,102 more rows, and 55 more variables: v126 <int>, v127 <int>, # v128 <int>, v129 <int>, v130 <int>, v131 <int>, v132 <int>, v133 <int>, # v134 <int>, v135 <int>, v136 <int>, v137 <int>, v138 <int>, v139 <int>, # v140 <int>, v141 <int>, v142 <int>, v143 <int>, v144 <int>, v145 <int>, # v146 <int>, v147 <int>, v148 <int>, v149 <int>, v150 <int>, v151 <int>, # v152 <int>, v153 <int>, v154 <int>, v155 <int>, v156 <int>, v157 <int>, # v158 <int>, v159 <int>, v160 <int>, v161 <int>, v162 <int>, v163 <int>, # v164 <int>, v165 <int>, v166 <int>, v167 <int>, v168 <int>, v169 <int>, # v170 <int>, v171 <int>, v172 <int>, v173 <int>, v174 <int>, v175 <int>, # v176 <int>, v177 <int>, v178 <int>, v179 <int>, v180 <int>
SLIDE 5
DataCamp Supervised Learning in R: Case Studies
SLIDE 6
DataCamp Supervised Learning in R: Case Studies
Opinions and attitudes in the 1960s
"Catholics should boycott indecent movies." "In the past 25 years, this country has moved dangerously close to socialism." "I would rather be called an idealist than a practical person."
SLIDE 7
DataCamp Supervised Learning in R: Case Studies
Tidy your data
> sisters67 %>% + select(-sister) %>% + gather(key, value, -age) # A tibble: 5,012,280 x 3 age key value <dbl> <chr> <int> 1 60.0 v116 1 2 70.0 v116 2 3 60.0 v116 1 4 60.0 v116 5 5 50.0 v116 2 6 40.0 v116 4 7 50.0 v116 5 8 40.0 v116 5 9 30.0 v116 2 10 30.0 v116 4 # ... with 5,012,270 more rows
SLIDE 8
DataCamp Supervised Learning in R: Case Studies
Let's practice!
SUPERVISED LEARNING IN R: CASE STUDIES
SLIDE 9
DataCamp Supervised Learning in R: Case Studies
Exploratory data analysis with tidy data
SUPERVISED LEARNING IN R: CASE STUDIES
Julia Silge
Data Scientist at Stack Overflow
SLIDE 10
DataCamp Supervised Learning in R: Case Studies
Counting agreement
> tidy_sisters %>% + count(value) # A tibble: 5 x 2 value n <int> <int> 1 1 1303555 2 2 844311 3 3 645401 4 4 1108859 5 5 1110154
SLIDE 11
DataCamp Supervised Learning in R: Case Studies
Overall agreement with age
> tidy_sisters %>% + group_by(age) %>% + summarise(value = mean(value)) # A tibble: 9 x 2 age value <dbl> <dbl> 1 20.0 2.86 2 30.0 2.81 3 40.0 2.83 4 50.0 2.94 5 60.0 3.10 6 70.0 3.26 7 80.0 3.42 8 90.0 3.51 9 100 3.60
SLIDE 12
DataCamp Supervised Learning in R: Case Studies
Agreement on questions by age
tidy_sisters %>% filter(key %in% paste0("v", 153:170)) %>% group_by(key, value) %>% summarise(age = mean(age)) %>% ggplot(aes(value, age, color = key)) + geom_line(alpha = 0.5, size = 1.5) + geom_point(size = 2) + facet_wrap(~key)
SLIDE 13
DataCamp Supervised Learning in R: Case Studies
SLIDE 14
DataCamp Supervised Learning in R: Case Studies
SLIDE 15
DataCamp Supervised Learning in R: Case Studies
Freedom of speech
"People who don't believe in God have as much right to freedom of speech as anyone else."
SLIDE 16
DataCamp Supervised Learning in R: Case Studies
SLIDE 17
DataCamp Supervised Learning in R: Case Studies
Conservatism and heritage
"I like conservatism because it represents a stand to preserve our glorious heritage."
SLIDE 18
DataCamp Supervised Learning in R: Case Studies
SLIDE 19
DataCamp Supervised Learning in R: Case Studies
Vietnam War
"Catholics as a group should consider active opposition to US participation in Vietnam."
SLIDE 20
DataCamp Supervised Learning in R: Case Studies
SLIDE 21
DataCamp Supervised Learning in R: Case Studies
SLIDE 22
DataCamp Supervised Learning in R: Case Studies
Let's explore!
SUPERVISED LEARNING IN R: CASE STUDIES
SLIDE 23
DataCamp Supervised Learning in R: Case Studies
Predicting age with supervised machine learning
SUPERVISED LEARNING IN R: CASE STUDIES
Julia Silge
Data Scientist at Stack Overflow
SLIDE 24
DataCamp Supervised Learning in R: Case Studies
Build models
"rpart" "xgbLinear" "gbm"
SLIDE 25
DataCamp Supervised Learning in R: Case Studies
Choosing between multiple models
## CART sisters_cart <- train(age ~ ., method = "rpart", data = training) ## xgboost sisters_rf <- train(age ~ ., method = "xgbLinear", data = training) ## gbm sisters_gbm <- train(age ~ ., method = "gbm", data = training)
SLIDE 26
DataCamp Supervised Learning in R: Case Studies
SLIDE 27
DataCamp Supervised Learning in R: Case Studies
Why three data partitions?
Don't overestimate how well your model is performing!
SLIDE 28
DataCamp Supervised Learning in R: Case Studies
Why three data partitions?
> validation %>% + mutate(prediction = predict(sisters_xg, validation)) %>% + rmse(truth = age, estimate = prediction) [1] 13.27101 > testing %>% + mutate(prediction = predict(sisters_xg, testing)) %>% + rmse(truth = age, estimate = prediction) [1] 13.36945
SLIDE 29
DataCamp Supervised Learning in R: Case Studies
Let's practice!
SUPERVISED LEARNING IN R: CASE STUDIES
SLIDE 30
DataCamp Supervised Learning in R: Case Studies
You made it!
SUPERVISED LEARNING IN R: CASE STUDIES
Julia Silge
Data Scientist at Stack Overflow
SLIDE 31
DataCamp Supervised Learning in R: Case Studies
Predicting age
> metrics(model_results, truth = age, estimate = CART) # A tibble: 1 x 2 rmse rsq <dbl> <dbl> 1 14.8 0.170 > metrics(model_results, truth = age, estimate = XBG) # A tibble: 1 x 2 rmse rsq <dbl> <dbl> 1 13.3 0.338 > metrics(model_results, truth = age, estimate = GBM) # A tibble: 1 x 2 rmse rsq <dbl> <dbl> 1 12.8 0.382
SLIDE 32
DataCamp Supervised Learning in R: Case Studies
Predicting age
Build your model with your training data Choose your model with your validation data Evaluate your model with your testing data
SLIDE 33
DataCamp Supervised Learning in R: Case Studies
Diverse data, powerful tools
Fuel efficiency of cars Developers working remotely in the Stack Overflow survey Voter turnout in 2016 Catholic nuns' ages based on beliefs and attitudes
SLIDE 34
DataCamp Supervised Learning in R: Case Studies
Practical machine learning
Dealing with class imbalance Improving performance with resampling (bootstrap, cross-validation)
SLIDE 35
DataCamp Supervised Learning in R: Case Studies
Practical machine learning
Dealing with class imbalance Improving performance with resampling (bootstrap, cross-validation) Hyperparameter tuning?
SLIDE 36
DataCamp Supervised Learning in R: Case Studies
Practical machine learning
Try out multiple modeling approaches for each new problem Overall, perform well gradient tree boosting and random forest
SLIDE 37
DataCamp Supervised Learning in R: Case Studies
Never skip exploratory data analysis
SLIDE 38
DataCamp Supervised Learning in R: Case Studies
Go train some models!
SUPERVISED LEARNING IN R: CASE STUDIES