摘要：導(dǎo)入數(shù)據(jù)預(yù)處理計算值從到對應(yīng)的平均畸變程度用求解距離平均畸變程度用肘部法則來確定最佳的值建模

cus_general = customer[["wm_poi_id","city_type","pre_book","aor_type","is_selfpick_poi","is_selfpick_trade_poi"]]
cus_ord = customer[["wm_poi_id","month_original_price","month_order_cnt","service_fee_30day","abnor_rate_30day"]]
cus = customer[["wm_poi_id","comment_1star","comment_5star","pic_comment_cnt"]]
cus = customer[["wm_poi_id","waybill_received_ratio","waybill_delivered_ratio","waybill_ontime_ratio","waybill_normal_arrived_delivery_total_interval_avg","waybill_normal_poi_push_interval_avg","waybill_normal_receive_interval_avg","waybill_normal_fetch_interval_avg","waybill_normal_delivery_interval_avg","waybill_delivery_ontime_ratio","loss_amt"]]
cus_all = customer[["wm_poi_id","c5","ol_time","primary_first_tag_id","city_level",
                    "month_original_price","month_order_cnt","service_fee_30day","abnor_cnt_30day",
                    "comment_1star","comment_5star","pic_comment_cnt",
                    "area_30day","waybill_grab_5mins_ratio","waybill_delivered_ratio","waybill_normal_arrived_delivery_total_interval_avg","waybill_normal_receive_interval_avg",
                    "call.call_cnt","call.call_cnt_ord","call.call_cnt_poi","call.call_cnt_oth"]]

from sklearn import preprocessing
cus = pd.DataFrame(preprocessing.scale(cus_general.iloc[:,1:6]))
cus = pd.DataFrame(preprocessing.scale(cus_ord.iloc[:,1:5]))
cus = pd.DataFrame(preprocessing.scale(cus_all.iloc[:,1:21]))
cus.columns = ["city_type","pre_book","aor_type","is_selfpick_poi","is_selfpick_trade_poi"]
cus.columns = ["month_original_price","month_order_cnt","service_fee_30day","abnor_rate_30day"]
cus.columns = ["comment_1star","comment_5star","pic_comment_cnt"]
cus.columns = ["waybill_push_ratio","waybill_delivered_ratio","waybill_ontime_ratio","waybill_normal_arrived_delivery_total_interval_avg","waybill_normal_poi_push_interval_avg","waybill_normal_receive_interval_avg","waybill_normal_fetch_interval_avg","waybill_normal_delivery_interval_avg","waybill_delivery_ontime_ratio","loss_amt"]
cus.columns = ["c5","ol_time","primary_first_tag_id","city_level",
               "month_original_price","month_order_cnt","service_fee_30day","abnor_cnt_30day",
               "comment_1star","comment_5star","pic_comment_cnt",
               "area_30day","waybill_grab_5mins_ratio","waybill_delivered_ratio","waybill_normal_arrived_delivery_total_interval_avg","waybill_normal_receive_interval_avg",
               "call.call_cnt","call.call_cnt_ord","call.call_cnt_poi","call.call_cnt_oth"]

from sklearn.cluster import KMeans
from scipy.spatial.distance import cdist
K=range(1,15)
meandistortions=[]
for k in K:
    kmeans=KMeans(n_clusters=k)
    kmeans.fit(cus)
    meandistortions.append(sum(np.min(cdist(cus,kmeans.cluster_centers_,"euclidean"),axis=1)))
plt.plot(K,meandistortions,"bx-")
plt.xlabel("k")
plt.ylabel(u"平均畸變程度")
plt.title(u"用肘部法則來確定最佳的K值")

from sklearn.cluster import KMeans
clf = KMeans(n_clusters=12)
clf.fit(cus)
pd.Series(pd.Series(clf.labels_).value_counts())

centres = pd.DataFrame(clf.cluster_centers_)
centres.columns = cus_all.iloc[:,1:21].columns
centres.plot(kind="bar", subplots=True, figsize=(6,15))
clf.inertia_

cus_general = pd.concat([cus_general, pd.DataFrame(clf.fit_predict(cus))], axis=0)
cus_general = cus_general.rename(columns={0:"general"})
cus_ord = pd.concat([cus_ord, pd.DataFrame(clf.fit_predict(cus))], axis=0)
cus_ord = cus_ord.rename(columns={0:"order"})
cus_all = pd.concat([cus_all, pd.DataFrame(clf.fit_predict(cus))], axis=0)
cus_all = cus_all.rename(columns={0:"cluster"})

centres = cus_all.groupby(["cluster"]).mean()

cus_all.to_csv("cluster.csv")

result = cus_all[cus_all["cluster"]==2]