結論: 最後, 發現真正管用的是PDB..
1. pdb能順利trace進入外部程式模組..
2. pdb.run("print(df)")
3. 還是找到了VS Code debug外部程式的關鍵: 真的是在justmycode=false後面加上purpose: ["debug-in-terminal"]就OK了
4. 不錯的debug功能: 預覽呼叫階層(可以反查某函數被誰呼叫)
[关于vscode调试时跳不到本.py或本.ipynb文件之外的问题_vscode打不开ipynb_Jay E的博客-CSDN博客](https://blog.csdn.net/Edward__J/article/details/129804740?utm_medium=distribute.wap_relevant.none-task-blog-2~default~baidujs_baidulandingword~default-1-129804740-blog-111991099.237&fbclid=IwAR1_J5RbBaO4iZEGaitTVtWXtW1RvtbfFstK2waBugQG8taZZg2m_Smw6pk) [[Obsidian-Highlights]]
問題:
2023-08-04 09:44:34.985121 首个交易日： 2023-05-29
2023-08-04 09:44:34.985121 最后交易日： 2023-05-27
2023-08-04 09:44:34.985121 总交易日： 64
2023-08-04 09:44:34.985121 盈利交易日： 35
2023-08-04 09:44:34.985121 亏损交易日： 29
找到問題的源頭了, df資料是錯的..看時間就知了..
df, 和data.csv不同, 差很多, 不知是否與其內部原有的sqlite相衝突?
檢查由pd.read_csv所讀入之df, 看來正常..
進一步檢查self.history_data, 看來也正常..
還是找到了VS Code debug外部程式的關鍵: 真的是在justmycode=false後面加上purpose: ["debug-in-terminal"]就OK了
找到出現怪異日期之處了: backtesting_data[0]
def run_backtesting(self) -> None:
""""""
if self.mode == BacktestingMode.BAR:
func = self.new_bar
else:
func = self.new_tick
self.strategy.on_init()
# Use the first [days] of history data for initializing strategy
day_count: int = 0
ix: int = 0
for ix, data in enumerate(self.history_data):
if self.datetime and data.datetime.day != self.datetime.day:
day_count += 1
if day_count >= self.days:
break
self.datetime = data.datetime
try:
self.callback(data)
except Exception:
self.output("触发异常，回测终止")
self.output(traceback.format_exc())
return
self.strategy.inited = True
self.output("策略初始化完成")
self.strategy.on_start()
self.strategy.trading = True
self.output("开始回放历史数据")
# Use the rest of history data for running backtesting
backtesting_data: list = self.history_data[ix:]
if len(backtesting_data) <= 1:
self.output("历史数据不足，回测终止")
return
total_size: int = len(backtesting_data)
batch_size: int = max(int(total_size / 10), 1)
for ix, i in enumerate(range(0, total_size, batch_size)):
batch_data: list = backtesting_data[i: i + batch_size]
for data in batch_data:
try:
func(data)
except Exception:
self.output("触发异常，回测终止")
self.output(traceback.format_exc())
return
progress = min(ix / 10, 1)
progress_bar: str = "=" * (ix + 1)
self.output(f"回放进度：{progress_bar} [{progress:.0%}]")
self.strategy.on_stop()
self.output("历史数据回放结束")
backtesting中之calculate_statistics中的df有錯..
def calculate_statistics(self, df: DataFrame = None, output=True) -> dict:
""""""
self.output("开始计算策略统计指标")
#breakpoint()
#import pdb; pdb.set_trace()
# Check DataFrame input exterior
if df is None:
df: DataFrame = self.daily_df
# Init all statistics default value
start_date: str = ""
end_date: str = ""
total_days: int = 0
profit_days: int = 0
loss_days: int = 0
end_balance: float = 0
self.daily_df.head(60)
close_price pre_close ... total_pnl net_pnl
date ...
2023-05-29 16567 1 ... -13800.0 -14009.5170
2023-05-30 16588 16567 ... -153360.0 -156289.8774
2023-05-31 16416 16588 ... 258000.0 258000.0000
2023-06-01 16571 16416 ... -147180.0 -148432.2744
2023-06-02 16749 16571 ... 53400.0 53400.0000
2023-06-03 16762 16749 ... 3900.0 3900.0000
2023-06-05 16693 16762 ... -53460.0 -54720.9252
2023-06-06 16798 16693 ... 31500.0 31500.0000
2023-06-07 16879 16798 ... 24300.0 24300.0000
2023-06-08 16814 16879 ... 27120.0 24583.9098
2023-06-09 16901 16814 ... 26100.0 26100.0000
2023-06-10 16930 16901 ... 8700.0 8700.0000
2023-06-12 17033 16930 ... 30900.0 30900.0000
2023-06-13 17249 17033 ... 64800.0 64800.0000
2023-06-14 17249 17249 ... -139140.0 -141719.8518
2023-06-15 17326 17249 ... -53580.0 -54873.7464
2023-06-16 17244 17326 ... -57000.0 -58290.2040
2023-06-17 17220 17244 ... 36000.0 36000.0000
2023-06-19 17235 17220 ... -22500.0 -22500.0000
2023-06-20 17145 17235 ... 135000.0 135000.0000
2023-06-21 16937 17145 ... -118020.0 -120588.7926
2023-06-22 16919 16937 ... 27000.0 27000.0000
2023-06-26 16865 16919 ... 81000.0 81000.0000
2023-06-27 16790 16865 ... 112500.0 112500.0000
2023-06-28 16787 16790 ... -86940.0 -88209.2412
2023-06-29 16749 16787 ... 136380.0 135111.1206
2023-06-30 16878 16749 ... -62460.0 -63727.4808
2023-07-01 16865 16878 ... -3900.0 -3900.0000
2023-07-03 17008 16865 ... 42900.0 42900.0000
2023-07-04 17105 17008 ... 29100.0 29100.0000
2023-07-05 16917 17105 ... 68700.0 67422.7290
2023-07-06 16549 16917 ... 552000.0 552000.0000
2023-07-07 16695 16549 ... -361920.0 -364434.3012
2023-07-08 16671 16695 ... -19440.0 -20701.8972
2023-07-10 16660 16671 ... -65400.0 -66657.7770
2023-07-11 16863 16660 ... 22740.0 21479.2152
2023-07-12 17034 16863 ... 51300.0 51300.0000
2023-07-13 17198 17034 ... -84360.0 -86932.5348
2023-07-14 17289 17198 ... 27300.0 27300.0000
2023-07-15 17243 17289 ... -13800.0 -13800.0000
2023-07-17 17287 17243 ... 13200.0 13200.0000
2023-07-18 17258 17287 ... -29400.0 -30691.3110
2023-07-19 17094 17258 ... 68880.0 66297.8748
2023-07-20 16921 17094 ... 235200.0 233914.7910
2023-07-21 16926 16921 ... -26940.0 -28214.7492
2023-07-22 16912 16926 ... 21000.0 21000.0000
2023-07-24 17070 16912 ... -99300.0 -100577.6490
2023-07-25 17244 17070 ... 52200.0 52200.0000
2023-07-26 17128 17244 ... -34980.0 -36264.9066
2023-07-27 17280 17128 ... -42780.0 -44067.5634
2023-07-28 17375 17280 ... -84360.0 -86940.7374
2023-07-29 17379 17375 ... 1200.0 1200.0000
2023-07-31 17113 17379 ... 224940.0 223646.7558
2023-08-01 17119 17113 ... -136260.0 -137548.2438
2023-05-17 15935 17119 ... -371400.0 -372213.1740
2023-05-18 16118 15935 ... -51900.0 -52716.8640
2023-05-19 16092 16118 ... -7800.0 -7800.0000
2023-05-20 16103 16092 ... 3300.0 3300.0000
2023-05-22 16124 16103 ... 6300.0 6300.0000
2023-05-23 16094 16124 ... -41760.0 -42578.4012
[60 rows x 13 columns]
不錯的debug功能: 預覽呼叫階層(可以反查某函數被誰呼叫)
即使進入了csvbacktesting->backtesting->run_backtesting時: self.history_data, 資料是正確的..
問題就在以下這段碼產生的backtesting_data有誤:
def run_backtesting(self) -> None:
""""""
if self.mode == BacktestingMode.BAR:
func = self.new_bar
else:
func = self.new_tick
self.strategy.on_init()
# Use the first [days] of history data for initializing strategy
day_count: int = 0
ix: int = 0
for ix, data in enumerate(self.history_data):
if self.datetime and data.datetime.day != self.datetime.day:
day_count += 1
if day_count >= self.days:
break
self.datetime = data.datetime
try:
self.callback(data)
except Exception:
self.output("触发异常，回测终止")
self.output(traceback.format_exc())
return
self.strategy.inited = True
self.output("策略初始化完成")
self.strategy.on_start()
self.strategy.trading = True
self.output("开始回放历史数据")
# Use the rest of history data for running backtesting
backtesting_data: list = self.history_data[ix:]
if len(backtesting_data) <= 1:
self.output("历史数据不足，回测终止")
return
total_size: int = len(backtesting_data)
batch_size: int = max(int(total_size / 10), 1)
for ix, i in enumerate(range(0, total_size, batch_size)):
batch_data: list = backtesting_data[i: i + batch_size]
for data in batch_data:
try:
func(data)
except Exception:
self.output("触发异常，回测终止")
self.output(traceback.format_exc())
return
progress = min(ix / 10, 1)
progress_bar: str = "=" * (ix + 1)
self.output(f"回放进度：{progress_bar} [{progress:.0%}]")
self.strategy.on_stop()
self.output("历史数据回放结束")
使用debug visualizer插件:
找出ix值是關鍵.. ix用途?
for ix, data in enumerate(self.history_data):
if self.datetime and data.datetime.day != self.datetime.day:
day_count += 1
if day_count >= self.days:
break
self.datetime = data.datetime
5. 日内对齐等交易时长K线生成器的实现
5.1 确定K线生成器MyBarGenerator的生成规则
5.1.1 一步到位地解决问题
先给它取个名称，就叫MyBarGenerator吧，它是对BarGenerator的扩展。
不过在构思MyBarGenerator的时候，我发现它其实不应该叫“日内对齐等交易时长K线生成器”。因为我们不应该只局限于日内的n分钟K线生成器，难道vnpy系统就不应该、不能够或者不使用日线以上的K线了吗？我们只能够使用日内K线进行量化交易吗？难道大家都没有过这方面的需求吗？我想答案是否定的。
那好，所幸就设计一个全功能的K线生成器：MyBarGenerator。
为此我们需要扩展Interval的定义，因为Interval是表示K线周期的常量，可是它的格局不够，最大只能到周一级WEEKLY。也就是说您用目前的Interval是没有办法表达月和年这样的周期的。
class Interval(Enum):
"""
Interval of bar data.
"""
MINUTE = "1m"
HOUR = "1h"
DAILY = "d"
WEEKLY = "w"
TICK = "tick"
MONTHLY = "month" # hxxjava add
YEARLY = "year" # hxxjava add顺便在这里吐槽一下BarGenerator：
- 目前的BarData中包含了一个interval字段的，可是它在BarGenerator的时候根本就没有使用过，而使用它本是信手拈来的事情，但是没有却没有使用。如果不信，你可以去看看用它产生出来的bar的内容。
- 另外本来还应该增加一个秒单位（SECONDLY = "1s"）的，这个单位其实对高频交易也是很有需求的，可是现在却没有。不知道大家对此有什么看法。
https://www.vnpy.com/forum/topic/30193-che-di-jie-jue-kxian-sheng-cheng-qi-de-wen-ti-yi-ge-ri-nei-dui-qi-deng-jiao-yi-shi-chang-de-kxian-sheng-cheng-qi
沒有留言:
張貼留言