Using R to explore the UCI mushroom dataset reveals excellent KNN prediction results.

mushrooms_explore_a.png

mushrooms_explore_b.png

mushrooms_script.R

# Load our data set.
mushroom = read.table("mushrooms.data", header = TRUE, sep = ",");

mushroom$class = as.character(mushroom$class);
mushroom$class[mushroom$class == "e"] = "edible";
mushroom$class[mushroom$class == "p"] = "poisonous";
mushroom$class = as.factor(mushroom$class);

# Remove incomplete data.
mushroom$veiltype = NULL;
mushroom$stalkroot = NULL;

# Load libraries.
library(ggplot2);
library(caret);
library(gridExtra);

set.seed(100);

# Inspect our data set.
names(mushroom);
head(mushroom);

# Explore the data.
qp1 = qplot(capcolor, habitat, color = class, data = mushroom, geom = "jitter", main = "Mushroom data set - Cap color vs Habitat");
qp2 = qplot(capcolor, habitat, color = class, data = mushroom, geom = "jitter", facets = bruises ~ ., main = "Mushroom data set by Bruises - Cap color vs Habitat");
qp3 = qplot(odor, sporeprintcolor, color = class, data = mushroom, geom = "jitter", main = "Mushroom data set - Odor vs Spore Print Color");
qp4 = qplot(stalksurfacebelowring, stalkcolorabovering, color = class, data = mushroom, geom = "jitter", main = "Mushroom data set - Stalk Surface below Ring vs Stalk Color above Ring");

grid.arrange(qp1, qp2, qp3, qp4, ncol = 2, nrow = 2);

# Explore the overlap where there is no odor and a spore print color of white.
mushroom$class = as.character(mushroom$class)
mushroom$class[mushroom$odor == "n" & mushroom$sporeprintcolor == "w" & mushroom$class == "poisonous"] = "poisonous2"
mushroom$class[mushroom$odor == "n" & mushroom$sporeprintcolor == "w" & mushroom$class == "edible"] = "edible2"
mushroom$class = as.factor(mushroom$class)

qp5 = qplot(capcolor, habitat, color = class, data = mushroom, geom = "jitter", main = "Mushroom data set - Cap color vs Habitat");
qp6 = qplot(capcolor, habitat, color = class, data = mushroom, geom = "jitter", facets = bruises ~ ., main = "Mushroom data set by Bruises - Cap color vs Habitat");
qp7 = qplot(odor, sporeprintcolor, color = class, data = mushroom, geom = "jitter", main = "Mushroom data set - Odor vs Spore Print Color");
qp8 = qplot(stalksurfacebelowring, stalkcolorabovering, color = class, data = mushroom, geom = "jitter", main = "Mushroom data set - Stalk Surface below Ring vs Stalk Color above Ring");

grid.arrange(qp5, qp6, qp7, qp8, ncol = 2, nrow = 2);

# Restore.
mushroom$class = as.character(mushroom$class);
mushroom$class[mushroom$class == "edible2"] = "edible";
mushroom$class[mushroom$class == "poisonous2"] = "poisonous";
mushroom$class = as.factor(mushroom$class);

# Split our data into a training (70%) and test (30%) set.
training_split = createDataPartition(y = mushroom$class, p = 0.70, list = FALSE);
training_set = mushroom[training_split,];
testing_set = mushroom[-training_split,];

model_fit = train(class ~ ., method = "knn", data = training_set, trControl = trainControl(method = 'cv', number = 10, classProbs = TRUE));
print(model_fit);
plot(model_fit);

# Classify from our reserved test set.
testing_set_predict = predict(model_fit, newdata = testing_set[, -1]); # Remove the Class field to prove we are not cheating.
# Verifying our model from the classifications.
table(testing_set_predict, testing_set$class);

testing_set$Correct = testing_set_predict == testing_set$class;
accuracy = length(testing_set$Correct[testing_set$Correct == TRUE]) / length(testing_set$Correct);
paste("Training accuracy:", accuracy);

important_features = varImp(model_fit, scale = FALSE);
print(important_features);
plot(important_features);

The visualization is very interesting. Can we do the same with python?
Also what did you learn from the data exploration? How did it help in classification?