Question

Can you create an example of a decision tree model with Python that features data processing, data mining, and result playing? You have to implement it from scratch, and no library databases can be used.

Answer 1

This is a code for the decision tree in which only default python packages are used.

from random import seed
from random import randrange
from csv import reader

#load a csv
def load_csv(fName):
file=open(fName,"rb")
lines=reader(file)
dataset=list(lines)
return dataset

# Convert string column to float

def str_column_to_float(dataSet, column):
   for row in dataSet:
       row[column] = float(row[column].strip())

# Split a dataset into k folds

def cross_validation_split(dataset,folds):
   dataset_split = list()
   dataset_copy = list(dataset)
   fold_size = int(len(dataset) / folds)
   for i in range(folds):
       fold = list()
       while len(fold) < fold_size:
           index = randrange(len(dataset_copy))
           fold.append(dataset_copy.pop(index))
       dataset_split.append(fold)
   return dataset_split

# Calculate accuracy percentage

def accuracy_metric(actual, predicted):
   correct = 0
   for i in range(len(actual)):
       if actual[i] == predicted[i]:
           correct += 1
   return correct / float(len(actual)) * 100.0

# Evaluate an algorithm using a cross validation split

def evaluate_algorithm(dataset, algorithm, fold_cnt, *args):
   folds = cross_validation_split(dataset, fold_cnt)
   scores = list()
   for fold in folds:
       train_set = list(folds)
       train_set.remove(fold)
       train_set = sum(train_set, [])
       test_set = list()
       for row in fold:
           row_copy = list(row)
           test_set.append(row_copy)
           row_copy[-1] = None
       predicted = algorithm(train_set, test_set, *args)
       actual = [row[-1] for row in fold]
       accuracy = accuracy_metric(actual, predicted)
       scores.append(accuracy)
   return scores

# Split a dataset based on an attribute and an attribute value

def test_split(index, value, dataSet):
   left, right = list(), list()
   for row in dataSet:
       if row[index] < value:
           left.append(row)
       else:
           right.append(row)
   return left, right

# Calculate the Gini index for a split dataset

def gini_index(groups, classes):
   # count all samples at split point
   n_instances = float(sum([len(group) for group in groups]))
   # sum weighted Gini index for each group
   gini = 0.0
   for group in groups:
       size = float(len(group))
       # avoid divide by zero
       if size == 0:
           continue
       score = 0.0
       # score the group based on the score for each class
       for class_val in classes:
           p = [row[-1] for row in group].count(class_val) / size
           score += p * p
       # weight the group score by its relative size
       gini += (1.0 - score) * (size / n_instances)
   return gini

# Select the best split point for a dataset

def get_split(dataSet):
   class_values = list(set(row[-1] for row in dataSet))
   b_index, b_value, b_score, b_groups = 999, 999, 999, None
   for index in range(len(dataSet[0])-1):
       for row in dataSet:
           groups = test_split(index, row[index], dataSet)
           gini = gini_index(groups, class_values)
           if gini < b_score:
               b_index, b_value, b_score, b_groups = index, row[index], gini, groups
   return {'index':b_index, 'value':b_value, 'groups':b_groups}

# Create a terminal node value

def to_terminal(grp):
   outcomes = [row[-1] for row in grp]
   return max(set(outcomes), key=outcomes.count)

# Create child splits for a node or make terminal

def split(node, max_depth, min_size, depth):
   left, right = node['groups']
   del(node['groups'])
   # check for a no split
   if not left or not right:
       node['left'] = node['right'] = to_terminal(left + right)
       return
   # check for max depth
   if depth >= max_depth:
       node['left'], node['right'] = to_terminal(left), to_terminal(right)
       return
   # process left child
   if len(left) <= min_size:
       node['left'] = to_terminal(left)
   else:
       node['left'] = get_split(left)
       split(node['left'], max_depth, min_size, depth+1)
   # process right child
   if len(right) <= min_size:
       node['right'] = to_terminal(right)
   else:
       node['right'] = get_split(right)
       split(node['right'], max_depth, min_size, depth+1)

# Build a decision tree

def build_tree(train, max_depth, min_size):
   root = get_split(train)
   split(root, max_depth, min_size, 1)
   return root

# Make a prediction with a decision tree

def predict(node, row):
   if row[node['index']] < node['value']:
       if isinstance(node['left'], dict):
           return predict(node['left'], row)
       else:
           return node['left']
   else:
       if isinstance(node['right'], dict):
           return predict(node['right'], row)
       else:
           return node['right']

# Classification and Regression Tree Algorithm
def decision_tree(train, test, max_depth, min_size):
   tree = build_tree(train, max_depth, min_size)
   predictions = list()
   for row in test:
       prediction = predict(tree, row)
       predictions.append(prediction)
   return(predictions)

# Test
seed(1)
# load and prepare data
filename = 'your_file_name.csv'
dataset = load_csv(filename)
# convert string attributes to integers
for i in range(len(dataset[0])):
   str_column_to_float(dataset, i)
# evaluate algorithm
n_folds = 5
max_depth = 5
min_size = 10
scores = evaluate_algorithm(dataset, decision_tree, n_folds, max_depth, min_size)
print('Scores: %s' % scores)
print('Mean Accuracy: %.3f%%' % (sum(scores)/float(len(scores))))