10秒後のレコードの情報を読み込む方法について

今、10秒後の株価の騰落率を返すプログラムを書いています。
とりあえず、for文を使って上手く動くコードは書けたのですが、実行に時間がかかりすぎるため、pandasの処理でうまく返せないか検討しています。

いいアイデアがある方、ご教示いただけないでしょうか？

python
1        START_TIME = datetime.time(9, 00, 0, 0)
2        END_TIME = datetime.time(15, 00, 0, 0)
3        RESTSTART_TIME = datetime.time(11, 30, 0, 0)
4        RESTEND_TIME = datetime.time(12, 30, 0, 0)
5
6        df = args
7        df = df.reset_index(drop=True)
8
9        for i, row in df.iterrows():
10            t = pushlist.objects.get(pushlist_id=df.loc[i, 'pushlist_id'])
11            time = df.loc[i, 'currentpricetime'] + datetime.timedelta(hours=9)
12            if time.time() > END_TIME:  # 教師信号となる時間が15時を超えていたらelseにとんで0をいれる
13                print("時間外")
14            else:
15                # ここから10秒後の教師信号
16                time2 = time + datetime.timedelta(seconds=10)
17                if RESTSTART_TIME < time2.time() < RESTEND_TIME:
18                    t.later10s = 99  # 99は時間外
19                else:
20                    time3 = time2 - datetime.timedelta(hours=9)
21                    df2 = df[(df.symbolname == df.loc[i, 'symbolname']) & (df.currentpricetime < time3)]  # 同じ銘柄かつ教師信号自転
22                    df2 = df2.sort_values(['currentpricetime'])
23                    df2 = df2.reset_index(drop=True)
24                    value = (float(df2['currentprice'][df2.index.max()]) / float(df['currentprice'][i])) - 1
25                    t.later10s = value
26                    t.save()
27

pandas
1	symbol	currentprice	currentpricetime	later10s	
24490	5170	2020-10-21 00:06:04.736356+00:00	0　←　00:06:14の株価との騰落率5160/5170-1 を返したい
34490	5170	2020-10-21 00:06:06.095071+00:00	0	
44490	5170	2020-10-21 00:06:06.271264+00:00	0		
54490	5160	2020-10-21 00:06:06.417457+00:00	0		
64490	5160	2020-10-21 00:06:06.469697+00:00	0		
74490	5160	2020-10-21 00:06:06.500777+00:00	0		
84490	5160	2020-10-21 00:06:06.946862+00:00	0		
94490	5160	2020-10-21 00:06:07.081324+00:00	0		
104490	5160	2020-10-21 00:06:07.912222+00:00	0		
114490	5160	2020-10-21 00:06:09.399203+00:00	0		
124490	5140	2020-10-21 00:06:10.346132+00:00	0		
134490	5140	2020-10-21 00:06:10.440464+00:00	0		
144490	5140	2020-10-21 00:06:10.505377+00:00	0		
154490	5140	2020-10-21 00:06:10.785968+00:00	0		
164490	5140	2020-10-21 00:06:10.982870+00:00	0		
174490	5140	2020-10-21 00:06:11.294839+00:00	0		
184490	5110	2020-10-21 00:06:11.837236+00:00	0		
194490	5110	2020-10-21 00:06:11.852856+00:00	0		
204490	5110	2020-10-21 00:06:12.022399+00:00	0		
214490	5110	2020-10-21 00:06:12.556094+00:00	0		
224490	5110	2020-10-21 00:06:12.805651+00:00	0		
234490	5130	2020-10-21 00:06:13.857057+00:00	0		
244490	5160	2020-10-21 00:06:13.888533+00:00	0		
254490	5160	2020-10-21 00:06:14.065261+00:00	0		
264490	5160	2020-10-21 00:06:14.220194+00:00	0		
274490	5160	2020-10-21 00:06:14.582616+00:00	0		
284490	5160	2020-10-21 00:06:15.189344+00:00	0		
294490	5160	2020-10-21 00:06:15.383874+00:00	0		
304490	5160	2020-10-21 00:06:17.014968+00:00	0		
314490	5140	2020-10-21 00:06:17.678799+00:00	0		
324490	5140	2020-10-21 00:06:17.730237+00:00	0		
334490	5140	2020-10-21 00:06:17.866179+00:00	0		
344490	5120	2020-10-21 00:06:20.887709+00:00	0		
354490	5120	2020-10-21 00:06:20.960669+00:00	0		
364490	5120	2020-10-21 00:06:21.106985+00:00	0		
374490	5120	2020-10-21 00:06:22.163820+00:00	0		
384490	5120	2020-10-21 00:06:23.252743+00:00	0		
394490	5120	2020-10-21 00:06:23.431432+00:00	0		
404490	5120	2020-10-21 00:06:23.600981+00:00	0	
414490	5120	2020-10-21 00:06:23.827950+00:00	0	
42

行動規範の内容に同意します

回答1件

ベストアンサー

多少は早くなると思います。
asofで10秒後(10秒以内の最終)の値をとってきています。

python
1# JSTに変換
2df['currentpricetime'] = df['currentpricetime'].dt.tz_convert('Asia/Tokyo')
3
4# symbolごとにわけて処理
5for symbol, df_s in df.groupby('symbol'):
6    # asof で10秒後(10秒以内の最終)のpriceを取得
7    later10s_price = df_s.set_index('currentpricetime')['currentprice'].asof(df_s['currentpricetime'] + pd.Timedelta('10s')).array
8    df.loc[df_s.index, 'later10s'] = later10s_price / df_s['currentprice'] - 1
9
10# 時間外の処理
11df.loc[df['currentpricetime'].dt.time > END_TIME, 'later10s'] = 0
12
13# 時間外の処理2
14time = (df['currentpricetime'] + pd.Timedelta('10s')).dt.time
15df.loc[(RESTSTART_TIME < time) & (time < RESTEND_TIME), 'later10s'] = 99