Este taller es para realizar problemas de Regression y Clasificación
Encuentre dos data set a utilizar:
Recomendación: Utilize los dataset disponisble en la librería scikit-learn http://scikit-learn.org/stable/datasets/
Cada uno escoja un par de dataset diferentes.
In [52]:
import numpy as np #libreria de datos numericos
import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd
import statsmodels.formula.api as smf
import pylab as pl
from sklearn import datasets
%matplotlib inline
from sklearn import datasets
In [2]:
data = pd.read_csv('http://www-bcf.usc.edu/~gareth/ISL/Advertising.csv', index_col= 0) # Data from the book: An introduction to Statistical learning
data.head()
Out[2]:
In [3]:
## SEABORN ##
sns.pairplot(data, x_vars=['TV','Radio','Newspaper'], y_vars='Sales', size=7, aspect=0.8)
# Seaborn library allows to display data, manipulate and analyze
Out[3]:
The TV vs. Sales plot shows that, aparently, there is a proportional increasement in Sales as the TV advertising is increassing. Therefore, it is suggested that Sales increass lineraly with the increassment of the TV advertising. The model wich can exlpain the behavior of the data is:
$y = \beta_0 + \beta_1x$
Together, $\beta_0$ and $\beta_1$ are called the model coefficients. However, $\beta_0$ is called the intercept. It means that $y$ reach $\beta_0$ when the others parameters are equal to 0
First of all, there are several ways to determine the values of the intercept and parameters of the model. 1) using STATSMODELS and 2) using SCIKIT-LEARN .
In [4]:
### STATSMODELS ###
# create a fitted model
lm1 = smf.ols(formula='Sales ~ TV', data=data).fit()
# print the coefficients
lm1.params
Out[4]:
In [11]:
### SCIKIT-LEARN ###
from sklearn.linear_model import LinearRegression
# create X and y
feature_cols = ['TV']
X = data[feature_cols]
y = data.Sales
# instantiate and fit
lm2 = LinearRegression()
lm2.fit(X, y)
# print the coefficients
print lm2.intercept_
print lm2.coef_
Now, let see if the data fits a lineal regression model.
In [76]:
sns.pairplot(data, x_vars=['TV','Radio','Newspaper'], y_vars='Sales', size=7, aspect=0.8, kind='reg')
# the parameter kind allow us not only to plot the liner regression model, but also to display the confidence intervals
Out[76]:
In [59]:
lm1.summary()
# This method from STATMODELS sumarize all the results of the regression
Out[59]:
From de summary it is posble to say that the model apparently explain the behavior of the data. Also, if it is chosen a value for X (it must betwen the extreme values of data) the model will predict the answer with 61 % of probability of being right. This is from the $ R^2 $ value. It is important to know that despite linear model has low variability it has a high bias because of the presumtion of data behavior.
In [79]:
from sklearn.datasets import load_iris
iris = load_iris()
type(iris) # special object bunch which is optimized to python and has DATA and TARGET ndarrays
Out[79]:
In [89]:
X1 = iris.data # store features (sepal and petal lenght and widht) in a matrix of 4 columns
y1 = iris.target # store what is called the response (target) in this case there are 3 responses (0,1,2) wich correspond to a target name
In [82]:
plt.scatter(X1[:, 0], X1[:, 1], c=y, cmap=plt.cm.Paired)
plt.xlabel('Sepal length')
plt.ylabel('Sepal width')
Out[82]:
In [83]:
plt.scatter(X1[:, 2], X1[:, 3], c=y1, cmap=plt.cm.Paired)
plt.xlabel('Petal length')
plt.ylabel('Petal width')
Out[83]:
There is a especial module (train_test_split), which allow us to split the datasets into two different. Each data set includes one set of the original set. However one of them is used to fit or train the model and the other one is used to make a prediction.
This separation is completly random in order to avoid bias. Nevertheless, it is posible to set a seed using the parameter 'random_state', so the partition will allways be the same.
In [104]:
from sklearn.cross_validation import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X1, y1, test_size= 0.4, random_state=4)
In [100]:
## 1. IMPORT THE CLASS/MODEL YOU PLAN TO USE ##
from sklearn.linear_model import LogisticRegression
## 2. Instantiate ## (Bring the model into an object or instance)
logreg = LogisticRegression()
## 3. Fit the model with data ##
logreg.fit(X_train,y_train)
## 4. Predict the response for a new observation
X_new = [[3,5,4,2],[5,4,3,2]]
logreg.predict(X_new)
Out[100]:
In [109]:
y_pred_log = logreg.predict(X_test)
from sklearn import metrics
X_test.shape
Out[109]:
In [110]:
print metrics.accuracy_score(y_test,y_pred_log)
In [114]:
from sklearn.neighbors import KNeighborsClassifier
## 2. Instantiate ## (Bring the model into an object or instance)
knn = KNeighborsClassifier() # by default this algorith compare an observation with the five nearest observation
## 3. Fit the model with data ##
knn.fit(X_train,y_train)
## 4. Predict the response for a new observation
y_pred_knn = knn.predict(X_test)
In [115]:
print metrics.accuracy_score(y_test,y_pred_knn)
In [118]:
knn.predict(X_new)
Out[118]:
In [117]:
knn_2 = KNeighborsClassifier(n_neighbors=6)
knn_2.fit(X_train,y_train)
y_pred_knn_2 = knn_2.predict(X_test)
print metrics.accuracy_score(y_test,y_pred_knn)
In [119]:
knn_2.predict(X_new)
Out[119]:
In [ ]: