Initial Test of tfestimators

tfestimators_test.R

Sys.setenv(PATH = '/Users/Sean/.virtualenvs/r-tensorflow/bin:$PATH')
library(tfestimators)
library(readr)
library(stringr)
library(purrr)

col_names = c(
  "age", "workclass", "fnlwgt", "education", "education_num",
  "marital_status", "occupation", "relationship", "race", "gender",
  "capital_gain", "capital_loss", "hours_per_week", "native_country",
  "income_bracket"
)

train <- read_csv("data/train_raw.csv", col_names = col_names, na = "?")
test  <- read_csv("data/test_raw.csv", col_names = col_names, na = "?", skip = 1)
test$income_bracket <- str_replace_all(test$income_bracket, fixed("."), "")

# input
infn <- function(data) {
  input_fn(data, 
    response = 'income_bracket',
    features = 'age'
  )
}

# feature engineering
age <- column_numeric("age")
age_buckets <- column_bucketized(age, boundaries = c(18, 25, 30, 35, 40, 45, 50, 55, 60, 65))
columns <- feature_columns(age_buckets)

# model 
model <- linear_classifier(
  feature_columns = columns,
  n_classes = 2,
  label_vocabulary = c("<=50K", ">50K")
)

#train 
model %>% train(infn(train))

# evaluate
res <- model %>% evaluate(infn(test))

# predict
pred <- model %>% predict(infn(test), predict_keys = prediction_keys()$CLASSES)

# flatten predictions, get confusion matrix
preds <- map_chr(pred, 'classes')
test$predicted <- preds

conf_matrix <- matrix(data = rep(0,4), nrow = 2, ncol = 2)
conf_matrix[1,1] <- sum(test$predicted == "<=50K" & test$income_bracket == "<=50K" )
conf_matrix[1,2] <- sum(test$predicted == ">50K" & test$income_bracket == "<=50K" )
conf_matrix[2,1] <- sum(test$predicted == "<=50K" & test$income_bracket == ">50K" )
conf_matrix[2,2] <- sum(test$predicted == ">50K" & test$income_bracket == ">50K" )
colnames(conf_matrix) <- c("Predicted <=50K", "Predicted >50K")
rownames(conf_matrix) <- c("Actual <=50K", "Actual >50K")
conf_matrix