PMR3508 - Aprendizado de Máquina e Reconhecimento de Padrões
Testando kNN com a base adult obtida no UCI repository. Iniciando com carregamento da base e com análise básica da base e dos atributos.
Autor: Fabio G. Cozman Data: 09/08/2018
In [1]:
import pandas as pd
import sklearn
In [2]:
adult = pd.read_csv("/Users/imac/Desktop/HOME/Didatico/Aulas/Graduacao/PMR3508/2018/Datasets/Adult-UCI/adult.data.txt",
names=[
"Age", "Workclass", "fnlwgt", "Education", "Education-Num", "Martial Status",
"Occupation", "Relationship", "Race", "Sex", "Capital Gain", "Capital Loss",
"Hours per week", "Country", "Target"],
sep=r'\s*,\s*',
engine='python',
na_values="?")
In [3]:
adult.shape
Out[3]:
In [4]:
adult.head()
Out[4]:
In [5]:
adult["Country"].value_counts()
Out[5]:
In [6]:
import matplotlib.pyplot as plt
In [7]:
adult["Age"].value_counts().plot(kind="bar")
Out[7]:
In [8]:
adult["Sex"].value_counts().plot(kind="bar")
Out[8]:
In [9]:
adult["Education"].value_counts().plot(kind="bar")
Out[9]:
In [10]:
adult["Occupation"].value_counts().plot(kind="bar")
Out[10]:
Retirando linhas com dados faltantes.
In [11]:
nadult = adult.dropna()
In [12]:
nadult
Out[12]:
Fazendo o mesmo processo com os dados de teste.
In [13]:
testAdult = pd.read_csv("/Users/imac/Desktop/HOME/Didatico/Aulas/Graduacao/PMR3508/2018/Datasets/Adult-UCI/adult.test.txt",
names=[
"Age", "Workclass", "fnlwgt", "Education", "Education-Num", "Martial Status",
"Occupation", "Relationship", "Race", "Sex", "Capital Gain", "Capital Loss",
"Hours per week", "Country", "Target"],
sep=r'\s*,\s*',
engine='python',
na_values="?")
In [14]:
nTestAdult = testAdult.dropna()
Primeiro teste: seleção de atributos numéricos, com kNN para k=3.
In [15]:
Xadult = nadult[["Age","Education-Num","Capital Gain", "Capital Loss", "Hours per week"]]
In [16]:
Yadult = nadult.Target
In [17]:
XtestAdult = nTestAdult[["Age","Education-Num","Capital Gain", "Capital Loss", "Hours per week"]]
In [18]:
YtestAdult = nTestAdult.Target
In [19]:
from sklearn.neighbors import KNeighborsClassifier
In [20]:
knn = KNeighborsClassifier(n_neighbors=3)
In [21]:
from sklearn.model_selection import cross_val_score
In [22]:
scores = cross_val_score(knn, Xadult, Yadult, cv=10)
In [23]:
scores
Out[23]:
In [24]:
knn.fit(Xadult,Yadult)
Out[24]:
In [25]:
YtestPred = knn.predict(XtestAdult)
In [26]:
YtestPred
Out[26]:
In [27]:
from sklearn.metrics import accuracy_score
In [28]:
accuracy_score(YtestAdult,YtestPred)
Out[28]:
Outro teste: mesmos dados, porém kNN com k=30. Melhor resultado obtido.
In [29]:
knn = KNeighborsClassifier(n_neighbors=30)
In [30]:
knn.fit(Xadult,Yadult)
Out[30]:
In [31]:
scores = cross_val_score(knn, Xadult, Yadult, cv=10)
In [32]:
scores
Out[32]:
In [33]:
YtestPred = knn.predict(XtestAdult)
In [34]:
accuracy_score(YtestAdult,YtestPred)
Out[34]:
Passando todos os dados não-numéricos para valores numéricos, e fazendo alguns testes com vários conjuntos de atributos (mantendo k=30, pois foi o valor de k que levou a melhor acurácia).
In [35]:
from sklearn import preprocessing
In [36]:
numAdult = nadult.apply(preprocessing.LabelEncoder().fit_transform)
In [37]:
numTestAdult = nTestAdult.apply(preprocessing.LabelEncoder().fit_transform)
In [38]:
Xadult = numAdult.iloc[:,0:14]
In [39]:
Yadult = numAdult.Target
In [40]:
XtestAdult = numTestAdult.iloc[:,0:14]
In [41]:
YtestAdult = numTestAdult.Target
In [42]:
knn = KNeighborsClassifier(n_neighbors=30)
In [43]:
knn.fit(Xadult,Yadult)
Out[43]:
In [44]:
YtestPred = knn.predict(XtestAdult)
In [45]:
accuracy_score(YtestAdult,YtestPred)
Out[45]:
In [46]:
Xadult = numAdult[["Age", "Workclass", "Education-Num", "Martial Status",
"Occupation", "Relationship", "Race", "Sex", "Capital Gain", "Capital Loss",
"Hours per week", "Country"]]
In [47]:
XtestAdult = numTestAdult[["Age", "Workclass", "Education-Num", "Martial Status",
"Occupation", "Relationship", "Race", "Sex", "Capital Gain", "Capital Loss",
"Hours per week", "Country"]]
In [48]:
knn = KNeighborsClassifier(n_neighbors=30)
In [49]:
knn.fit(Xadult,Yadult)
Out[49]:
In [50]:
YtestPred = knn.predict(XtestAdult)
In [51]:
accuracy_score(YtestAdult,YtestPred)
Out[51]:
In [52]:
Xadult = numAdult[["Age", "Workclass", "Education-Num",
"Occupation", "Race", "Sex", "Capital Gain", "Capital Loss",
"Hours per week"]]
In [53]:
XtestAdult = numTestAdult[["Age", "Workclass", "Education-Num",
"Occupation", "Race", "Sex", "Capital Gain", "Capital Loss",
"Hours per week"]]
In [54]:
knn = KNeighborsClassifier(n_neighbors=30)
In [55]:
knn.fit(Xadult,Yadult)
Out[55]:
In [56]:
YtestPred = knn.predict(XtestAdult)
In [57]:
accuracy_score(YtestAdult,YtestPred)
Out[57]: