tirthajyoti / mldsutils

My own ml and ds utils package

mldsutils: A machine-learning and data science utility package

Install

pip install mldsutils

How to use mldsutils.mldsutils

• Import the library
• Define a list of Scikit-learn estimators with your choice of hyperparameters
• Generate some synthetic data
• Run the run_regressor function to iterate through each of them and evaluate the given datatset

Import

from mldsutils.mldsutils import *
import numpy as np
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.datasets import make_classification

Classifiers and their names

classifiers = [KNeighborsClassifier(3),
               SVC(kernel="linear", C=0.025),
               SVC(gamma=2, C=10),]

clf_names = ['k-Nearest Neighbors(3)',
             'Support Vector Machine with Linear Kernel',
            'Support Vector Machine with RBF Kernel']

Some data

X1, y1 = make_classification(n_features=20, n_samples=2000,n_redundant=0, n_informative=20,
                             n_clusters_per_class=1,class_sep=0.5)

Run

Note, you will get back a Pandas DataFrame from this

d1,d2 = run_classifiers(X1,y1,
                        clf_lst=classifiers,names = clf_names,
                        runtime=True,
                        metric='f1',verbose=True)

Finished 10 runs for k-Nearest Neighbors(3) algorithm
---------------------------------------------------------------------------
Finished 10 runs for Support Vector Machine with Linear Kernel algorithm
---------------------------------------------------------------------------
Finished 10 runs for Support Vector Machine with RBF Kernel algorithm
---------------------------------------------------------------------------

Examining the result

Checking the dataframe of F1-scores

d1

	k-Nearest Neighbors(3)	Support Vector Machine with Linear Kernel	Support Vector Machine with RBF Kernel
0	0.944724	0.861314	0.000000
1	0.968215	0.855746	0.641766
2	0.946292	0.863850	0.289474
3	0.967742	0.856481	0.000000
4	0.969849	0.865116	0.020202
5	0.966234	0.875912	0.000000
6	0.967581	0.868545	0.000000
7	0.957179	0.852381	0.000000
8	0.962791	0.871194	0.000000
9	0.948980	0.897959	0.010204

Stats of the d1 to compare algorithms

d1.describe().T

	count	mean	std	min	25%	50%	75%	max
k-Nearest Neighbors(3)	10.0	0.959959	0.009884	0.944724	0.951029	0.964512	0.967702	0.969849
Support Vector Machine with Linear Kernel	10.0	0.866850	0.013142	0.852381	0.857690	0.864483	0.870532	0.897959
Support Vector Machine with RBF Kernel	10.0	0.096165	0.211790	0.000000	0.000000	0.000000	0.017703	0.641766

Checking the dataframe of fitting/training time

d2

	k-Nearest Neighbors(3)	Support Vector Machine with Linear Kernel	Support Vector Machine with RBF Kernel
0	0.979	43.942	124.595
1	0.000	45.477	137.699
2	0.000	45.893	107.373
3	0.981	49.376	141.589
4	0.000	44.429	148.940
5	0.976	43.940	147.467
6	0.977	42.083	134.906
7	0.974	46.962	139.639
8	0.000	42.484	130.852
9	0.976	61.050	113.273

d2.describe().T

	count	mean	std	min	25%	50%	75%	max
k-Nearest Neighbors(3)	10.0	0.5863	0.504610	0.000	0.00000	0.9750	0.97675	0.981
Support Vector Machine with Linear Kernel	10.0	46.5636	5.523546	42.083	43.94050	44.9530	46.69475	61.050
Support Vector Machine with RBF Kernel	10.0	132.6333	13.851391	107.373	126.15925	136.3025	141.10150	148.940

Visualizing the results with the plot_bars function

Make sure to pass the correct titles of the plots. Otherwise, default strings will be plotted which may indicate wrong thing for your experiment.

plot_bars(d1,t1="Mean F1 score of algorithms",
              t2="Std.dev of the F1 scores of algorithms")

plot_bars(d2,t1="Mean fitting time of algorithms",
              t2="Std.dev of the fitting time of algorithms")

How to use mldsutils.plots

from sklearn.neighbors import KNeighborsClassifier
from sklearn.datasets import make_classification
import matplotlib.pyplot as plt
from mldsutils.plots import *
        
X1, y1 = make_classification(n_features=10, n_samples=100,
                                     n_redundant=0, n_informative=10,
                                     n_clusters_per_class=1,class_sep=0.5)

plot_decision_boundaries(X1,y1,KNeighborsClassifier,n_neighbors=5)
plt.show()