1. machine-learning
  2. Regresion_logistica.ipynb

Ejercicio Python de Regresión Logística

Realizaremos un ejercicio de prueba para comprender como funciona este algoritmo

In [1]:
import pandas as pd
import numpy as np
from sklearn import linear_model
from sklearn import model_selection
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
import matplotlib.pyplot as plt
import seaborn as sb
%matplotlib inline

Cargamos los datos de entrada del archivo csv

In [2]:
dataframe = pd.read_csv(r"usuarios_win_mac_lin.csv")
dataframe.head()
Out[2]:
duracion paginas acciones valor clase
0 7.0 2 4 8 2
1 21.0 2 6 6 2
2 57.0 2 4 4 2
3 101.0 3 6 12 2
4 109.0 2 6 12 2
In [3]:
dataframe.describe()
Out[3]:
duracion paginas acciones valor clase
count 170.000000 170.000000 170.000000 170.000000 170.000000
mean 111.075729 2.041176 8.723529 32.676471 0.752941
std 202.453200 1.500911 9.136054 44.751993 0.841327
min 1.000000 1.000000 1.000000 1.000000 0.000000
25% 11.000000 1.000000 3.000000 8.000000 0.000000
50% 13.000000 2.000000 6.000000 20.000000 0.000000
75% 108.000000 2.000000 10.000000 36.000000 2.000000
max 898.000000 9.000000 63.000000 378.000000 2.000000
In [4]:
print(dataframe.groupby('clase').size())

clase
0    86
1    40
2    44
dtype: int64

Visualizamos los datos

In [5]:
dataframe.drop(['clase'],1).hist()
plt.show()
In [6]:
sb.pairplot(dataframe.dropna(), hue='clase',size=4,vars=["duracion", "paginas","acciones","valor"],kind='reg')
Out[6]:
<seaborn.axisgrid.PairGrid at 0x115804e50>

Creamos el modelo

In [7]:
X = np.array(dataframe.drop(['clase'],1))
y = np.array(dataframe['clase'])
X.shape
Out[7]:
(170, 4)
In [8]:
model = linear_model.LogisticRegression()
model.fit(X,y)
Out[8]:
LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
          intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,
          penalty='l2', random_state=None, solver='liblinear', tol=0.0001,
          verbose=0, warm_start=False)
In [9]:
predictions = model.predict(X)
print(predictions[0:5])

[2 2 2 2 2]
In [10]:
model.score(X,y)
Out[10]:
0.77647058823529413

Adicional: Validación del Modelo

In [11]:
validation_size = 0.20
seed = 7
X_train, X_validation, Y_train, Y_validation = model_selection.train_test_split(X, y, test_size=validation_size, random_state=seed)
In [12]:
name='Logistic Regression'
kfold = model_selection.KFold(n_splits=10, random_state=seed)
cv_results = model_selection.cross_val_score(model, X_train, Y_train, cv=kfold, scoring='accuracy')
msg = "%s: %f (%f)" % (name, cv_results.mean(), cv_results.std())
print(msg)

Logistic Regression: 0.743407 (0.115752)
In [13]:
predictions = model.predict(X_validation)
print(accuracy_score(Y_validation, predictions))

0.852941176471

Reporte de Resultados

In [14]:
print(confusion_matrix(Y_validation, predictions))

[[16  0  2]
 [ 3  3  0]
 [ 0  0 10]]
In [15]:
print(classification_report(Y_validation, predictions))

             precision    recall  f1-score   support

          0       0.84      0.89      0.86        18
          1       1.00      0.50      0.67         6
          2       0.83      1.00      0.91        10

avg / total       0.87      0.85      0.84        34

Clasificación de nuevos registros

In [16]:
X_new = pd.DataFrame({'duracion': [10], 'paginas': [3], 'acciones': [5], 'valor': [9]})
model.predict(X_new)
Out[16]:
array([2])