def get_avg_prop_list():
proportion_list = gen_proportion.get_proportion()
# 0-815 是前17天的
# 周末 ([5, 6, 12, 13] - 1)*48+(0……47)
# 工作日 ([1, 2, 3, 4, 7, 8, 9, 10, 11, 14, 15, 16, 17] - 1)*48+(0……47)
weekend_prop_list = []
weekend = [4, 5, 11, 12]
for i in range(48):
tmp = np.zeros((59, 59))
for j in weekend:
pos = j * 48 + i
tmp += proportion_list[pos]
tmp = tmp / 4
weekend_prop_list.append(tmp)
weekday_prop_list = []
weekday = np.array([0, 1, 2, 3, 6, 7, 8, 9, 10, 13, 14, 15, 16])
for i in range(48):
tmp = np.zeros((59, 59))
for j in weekday:
pos = j * 48 + i
tmp += proportion_list[pos]
tmp = tmp / 13
weekday_prop_list.append(tmp)
return weekday_prop_list, weekend_prop_list
def get_all_data():
# 读取测试集并加入counter
df_test = pd.read_csv('E:\\data\\DiDiData\\data_csv\\dataset\\Test.csv')
df_test['counter'] = (df_test['day']-1) * 48 + df_test['time'] + 1
df_weather = pd.read_csv('E:\\data\\DiDiData\\data_csv\\features\\before\\weather_feature.csv')
df_time = pd.read_csv('E:\\data\\DiDiData\\data_csv\\features\\time_feature.csv')
del df_time['date']
del df_time['time']
df = pd.concat([df_weather, df_time], axis=1)
df['counter'] = range(1, 1153)
# flow_total = list(df['count'].values)[816:]
df_tatol_predict = pd.read_csv('E:\\data\\DiDiData\\data_csv\\result\\gbrt_toal_result.csv')
flow_total = list(df_tatol_predict['count'].values)
proportion_list = gen_proportion.get_proportion()
return df, flow_total, proportion_list, df_test
def get_all_data():
try:
df_data = pd.read_csv(
'E:\\data\\DiDiData\\data_csv\\dataset\\data_3_week.csv')
except FileNotFoundError:
# 读取Train 和 Test 进行拼接，按照 date、time、start_district_id、dest_district_id 进行排序，并加入counter
df_train = pd.read_csv(
'E:\\data\\DiDiData\\data_csv\\dataset\\Train.csv')
df_test = pd.read_csv(
'E:\\data\\DiDiData\\data_csv\\dataset\\Test.csv')
df_data = pd.concat([df_train, df_test.loc[0:99123]], ignore_index=True)\
.sort_values(by=['date', 'time', 'start_district_id', 'dest_district_id'], axis=0, ascending=True)\
.reset_index(drop=True)
df_data['counter'] = (df_data['day'] - 1) * 48 + df_data['time'] + 1
df_data.set_index(
keys=['date', 'time', 'start_district_id', 'dest_district_id'],
inplace=True)
df_data.to_csv(
'E:\\data\\DiDiData\\data_csv\\dataset\\data_3_week.csv')
df_weather = pd.read_csv(
'E:\\data\\DiDiData\\data_csv\\features\\before\\weather_feature.csv')
df_time = pd.read_csv(
'E:\\data\\DiDiData\\data_csv\\features\\time_feature.csv')
del df_time['date']
del df_time['time']
df_feature = pd.concat([df_weather, df_time], axis=1)
df_feature['counter'] = range(1, 1153)
flow_total = list(df_feature['count'].values)
df_test = df_data.loc[319127:].reset_index(drop=True)
flow_total_m = []
for i in range(len(df_test)):
flow_total_m.append(df_feature.loc[df_test.loc[i, 'counter'] - 1,
'count'])
proportion_list = gen_proportion.get_proportion()
return df_data, df_test, df_feature, flow_total, flow_total_m, proportion_list,
x = np.arange(1, len(prop_list))
y = np.array(dist_avg)
print(y)
plt.title('最后一天与之前的平均矩阵距离')
plt.xlabel('天')
plt.xlim((0, len(day_list)))
plt.ylabel('平均比例矩阵距离')
plt.plot(x, y, color='red', marker='o', linestyle='-', linewidth=2)
plt.legend(loc='best')
plt.show()
def distance(prop1, prop2):
prop1 = np.array(prop1)
prop2 = np.array(prop2)
d = np.sqrt(np.sum(np.square(prop1 - prop2)))
return d
if __name__ == '__main__':
print('start time:', datetime.now().strftime('%Y-%m-%d %H:%M:%S'))
proportion_list = gp.get_proportion()
# measure_difference_day()
# measure_difference(24)
measure_difference_day_avg()
print('end time:', datetime.now().strftime('%Y-%m-%d %H:%M:%S'))