kaggle研究生招生（下）

对于该数据先采用回归算法，再转为分类算法，这次使用聚类算法

聚类算法（无监督机器学习算法）

df = pd.read_csv("../input/Admission_Predict.csv",sep = ",")
df=df.rename(columns = {'Chance of Admit ':'ChanceOfAdmit'})
serial = df["Serial No."]
df.drop(["Serial No."],axis=1,inplace = True)
df = (df- np.min(df))/(np.max(df)-np.min(df))
y = df.ChanceOfAdmit 
x = df.drop(["ChanceOfAdmit"],axis=1)

所有特征（x）均采用主成分分析法收集在一个特征中。

PCA

# for data visualization
from sklearn.decomposition import PCA
pca = PCA(n_components = 1, whiten= True )  # whitten = normalize
pca.fit(x)
x_pca = pca.transform(x)
x_pca = x_pca.reshape(400,)
dictionary = {"x":x_pca,"y":y}
data = pd.DataFrame(dictionary)
print("data:")
print(data.head())
print("\ndf:")
print(df.head())

K-means

利用肘形法确定K均值聚类的最佳聚类数是3

df["Serial No."] = serial
from sklearn.cluster import KMeans
wcss = []
for k in range(1,15):
    kmeans = KMeans(n_clusters=k)
    kmeans.fit(x)
    wcss.append(kmeans.inertia_)
plt.plot(range(1,15),wcss)
plt.xlabel("k values")
plt.ylabel("WCSS")
plt.show()

kmeans = KMeans(n_clusters=3)
clusters_knn = kmeans.fit_predict(x)

df["label_kmeans"] = clusters_knn


plt.scatter(df[df.label_kmeans == 0 ]["Serial No."],df[df.label_kmeans == 0].ChanceOfAdmit,color = "red")
plt.scatter(df[df.label_kmeans == 1 ]["Serial No."],df[df.label_kmeans == 1].ChanceOfAdmit,color = "blue")
plt.scatter(df[df.label_kmeans == 2 ]["Serial No."],df[df.label_kmeans == 2].ChanceOfAdmit,color = "green")
plt.title("K-means Clustering")
plt.xlabel("Candidates")
plt.ylabel("Chance of Admit")
plt.show()

df["label_kmeans"] = clusters_knn
plt.scatter(data.x[df.label_kmeans == 0 ],data[df.label_kmeans == 0].y,color = "red")
plt.scatter(data.x[df.label_kmeans == 1 ],data[df.label_kmeans == 1].y,color = "blue")
plt.scatter(data.x[df.label_kmeans == 2 ],data[df.label_kmeans == 2].y,color = "green")
plt.title("K-means Clustering")
plt.xlabel("X")
plt.ylabel("Chance of Admit")
plt.show()

层次聚类

采用树形图方法确定了层次聚类的最佳聚类数又是3

df["Serial No."] = serial

from scipy.cluster.hierarchy import linkage, dendrogram
merg = linkage(x,method="ward")
dendrogram(merg,leaf_rotation = 90)
plt.xlabel("data points")
plt.ylabel("euclidean distance")
plt.show()

from sklearn.cluster import AgglomerativeClustering
hiyerartical_cluster = AgglomerativeClustering(n_clusters = 3,affinity= "euclidean",linkage = "ward")
clusters_hiyerartical = hiyerartical_cluster.fit_predict(x)

df["label_hiyerartical"] = clusters_hiyerartical

plt.scatter(df[df.label_hiyerartical == 0 ]["Serial No."],df[df.label_hiyerartical == 0].ChanceOfAdmit,color = "red")
plt.scatter(df[df.label_hiyerartical == 1 ]["Serial No."],df[df.label_hiyerartical == 1].ChanceOfAdmit,color = "blue")
plt.scatter(df[df.label_hiyerartical == 2 ]["Serial No."],df[df.label_hiyerartical == 2].ChanceOfAdmit,color = "green")
plt.title("Hierarchical Clustering")
plt.xlabel("Candidates")
plt.ylabel("Chance of Admit")
plt.show()

plt.scatter(data[df.label_hiyerartical == 0 ].x,data.y[df.label_hiyerartical == 0],color = "red")
plt.scatter(data[df.label_hiyerartical == 1 ].x,data.y[df.label_hiyerartical == 1],color = "blue")
plt.scatter(data[df.label_hiyerartical == 2 ].x,data.y[df.label_hiyerartical == 2],color = "green")
plt.title("Hierarchical Clustering")
plt.xlabel("X")
plt.ylabel("Chance of Admit")
plt.show()

聚类算法比较

k-均值聚类和层次聚类是相似的。

总结

• 任何一个数据都可以进行回归，分类，甚至聚类的算法，关键如何处理数据采用不同种类的算法