Analyzing Iris Dataset¶
Prepared by Mahsa Sadi on 2020 - 06 - 22
In this notebook, we perform three steps:
- Reading the iris dataset.
- Visualizing the iris dataset.
- Building different models over the dataset and evaluate and compare their accuracy.
The iris data set contains data about different instances of three categories of iris flowers, namely setosa, versicolor and virginica. This data set measures four features (i.e.; attributes of the iris flowers, namely the length and width of sepal, and the length and width of petal) for different instances of iris flowers and identfies the category of each instance.
In [1]:
# pandas is a python library for manipulating and analyzing numerical tables and time-series
import pandas
print ("pandas {}".format (pandas.__version__))
from pandas.plotting import scatter_matrix
In [2]:
# matplotlib is a Python plotting library for numerical data
import matplotlib
print ("matplotlib {}".format (matplotlib.__version__))
import matplotlib.pyplot
In [3]:
# sklearn is a Python library for machine learning.
#sklearn contains various datasets and the reday-to-use implementation of various machine learning algorithms.
import sklearn
print ("sklearn {}".format (sklearn.__version__))
from sklearn import model_selection
In [4]:
from sklearn.metrics import classification_report
In [5]:
from sklearn.metrics import confusion_matrix
In [6]:
from sklearn.metrics import accuracy_score
In [7]:
from sklearn.linear_model import LogisticRegression
In [8]:
from sklearn.tree import DecisionTreeClassifier
In [9]:
from sklearn.neighbors import KNeighborsClassifier
In [10]:
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
In [11]:
from sklearn.naive_bayes import GaussianNB
In [12]:
from sklearn.svm import SVC
Reading Data¶
In [13]:
# The url of the iris dataset.
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data"
In [14]:
# The iris dataset has five columns (four of them are independant variables and one of them are dependant variable.)
names = ['sepal_length', 'sepal_width', 'petal_length', 'petal_width', 'class']
In [15]:
#Read the iris dataset from the defined url, the names of the columns in the datas set is defined in the names list
data_set = pandas.read_csv (url, names = names)
In [16]:
# Dispaly how many rows and columns the iris data set has
print (data_set.shape)
In [17]:
# Display the first 20 rows of the iris dataset
print (data_set.head (20))
Summerizing Data¶
In [18]:
# Provide a summary of the iris dataset
print (data_set.describe())
In [19]:
# See how many instances of each class exist
print (data_set.groupby('class').size())
Visualizing Data¶
In [21]:
#Draw a box plot for the iris dataset without sub plots
# A box plot helps gain insight about the distribution of each of feature (i.e.; attribute)
data_set.plot (kind = 'box', subplots = False, sharex = False, sharey = False)
Out[21]:
In [22]:
# Gain a summerized insight about the distribution of each feature in dataset
# Draw a box plot for the iris dataset with sub plots
# A box plot helps gain insight about the distribution of each of feature (i.e.; attribute)
data_set.plot (kind = 'box', subplots = True, layout = (2,2), sharex = False, sharey = False)
Out[22]:
In [23]:
# Gain detailed insight about the distribution of each feature in the dataset; i.e.; look carefully into the distribution of each feature
# Draw the histogram of the dataset.
# The histogram of the features helps us understand whether the distribution of a feature is normal or gGaussian or not.
data_set.hist ()
# From the following histograms we see that the length of petal and the width of sepal has almost a Gaussian distribution.
Out[23]:
In [24]:
# 4- Visualize Data
# Gain insight about the relationships between features
scatter_matrix (data_set)
Out[24]:
In [25]:
data_table = data_set. values
print (data_table [0:20])
Modeling and Analyzing Data¶
In [26]:
#independant variables of the iris flowers (the features)
X = data_table [:, 0:4]
#dependant variable (the category of the iris flowers )
Y = data_table [:, 4]
In [27]:
# Split the iris dataset into two sets: Training Set and Test Set
# 80% of the iris dataset is for training
#20% of the iris dataset is for testing
#the size of the test set
test_set_size = 0.2
#Randomly select train and test data set from the iris dataset
seed = 6
In [28]:
#Split the iris data set into training and test data sets and choose the training and test data set randomly.
X_train, X_test, Y_train, Y_test = model_selection.train_test_split (X,Y, test_size = test_set_size, random_state = seed)
In [29]:
# Use accuracy as the measure of the performance of the built model
scoring = 'accuracy'
In [30]:
# Build different models for the iris data
models = []
models.append (('Logistic Regression', LogisticRegression(solver ='lbfgs', multi_class = 'ovr')))
models.append (('Linear Discriminant Analysis', LinearDiscriminantAnalysis()))
models.append (('K Nearest Neigbors', KNeighborsClassifier()))
models.append (('CART', DecisionTreeClassifier()))
models.append (('Support Vector Machine', SVC(gamma ='scale')))
models.append (('Guassian Naive Bayes', GaussianNB()))
In [31]:
results = []
names = []
In [32]:
# When building different models, use cross validation by dividing the dataset into 10 slices, picking 8 of them for training set, and picking 2 of them for test set.
for name, model in models:
names.append (name)
K_Fold = model_selection.KFold (n_splits = 10, random_state = seed)
cv_results = model_selection.cross_val_score (model, X_train, Y_train, cv = K_Fold, scoring = scoring)
results.append (cv_results)
message = "%s: %f (%f)" % (name, cv_results.mean (), cv_results.std())
print (message)