首先，看看让人眼晕的一段代码，而且作者自己抱怨，非常耗时。 ``` # 选出打板的股票 def pick_high_limit(stocks,context,end_date,pre_date,pre_date_10): log.info('【pick_high_limit】') # 耗时 df_panel = get_price(stocks, count = 1,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','pre_close','low']) df_close = df_panel['close'] # log.info(df_close) df_open = df_panel['open'] df_high_limit = df_panel['high_limit'] df_low = df_panel['low'] df_money = df_panel['money'] df_pre_close = df_panel['pre_close'] high_limit_stock = [] for stock in (stocks): if(stock[0:2] == '30' or stock[0:2] == '68'): continue _high_limit = (df_high_limit[stock].values) _close = (df_close[stock].values) _open = (df_open[stock].values) _pre_close = (df_pre_close[stock].values) _low = (df_low[stock].values) if _close == _high_limit and _high_limit > _pre_close * 1.06: df_panel_50 = get_price(stock, count = 60,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) sum_plus_max_50 = (df_panel_50.loc[:,'close'] == df_panel_50.loc[:,'high_limit']).sum() # mean_60 = df_panel_50.loc[:,'close'].mean() # df_panel_30 = get_price(stock, count = 30,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) # mean_30 = df_panel_30.loc[:,'close'].mean() # df_panel_20 = get_price(stock, count = 20,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) # mean_20 = df_panel_20.loc[:,'close'].mean() df_panel_5 = get_price(stock, count = 5,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) sum_plus_max_5 = (df_panel_5.loc[:,'close'] == df_panel_5.loc[:,'high_limit']).sum() # df_panel_500 = get_price(stock, count = 500,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) # sum_plus_max_500 = (df_panel_500.loc[:,'close'] == df_panel_500.loc[:,'high_limit']).sum() # low_allday_500 = df_panel_500.loc[:,"close"].min() # high_alldayy_500 = df_panel_500.loc[:,"close"].max() # rate_500 = (high_alldayy_500 - low_allday_500) / low_allday_500 # df_panel_10 = get_price(stock, count = 10,end_date=end_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) # sum_plus_max_10 = (df_panel_10.loc[:,'close'] == df_panel_10.loc[:,'high_limit']).sum() df_panel_3 = get_price(stock, count = 2,end_date=pre_date, frequency='daily', fields=['open', 'close','high_limit','money','low'],skip_paused=True) sum_plus_max_3 = (df_panel_3.loc[:,'close'] == df_panel_3.loc[:,'high_limit']).sum() # mean_10 * 0.0455 > mean_20 * 0.0181 and mean_20 * 0.0181 > mean_30 * 0.0158 and mean_30 * 0.0158 > mean_60 * 0.0094 and rate_500 < 2 and sum_plus_max_50 < = 4 and if sum_plus_max_3 == 0 and sum_plus_max_5 < = 2: high_limit_stock.append(stock) return high_limit_stock ``` 那怎么去处理这看的人头晕目眩的代码？ ### 一、去掉那些被注释掉的语句 1.1 在Pycharm中，可以通过正则表达式`^\s*#.*$`，去查找替换成空行。 1.2 Reformat file（Ctrl+Shift+L）重新格式化Py文件 ### 二、去掉无用的赋值语句 无用的赋值语句，在Pycharm里面，看的非常清楚， ``` df_money = df_panel['money'] ...... sum_plus_max_50 = (df_panel_50.loc[:, 'close'] == df_panel_50.loc[:, 'high_limit']).sum() ...... ``` 当删除掉`sum_plus_max_50`这一条语句后，会发现，它前面的一天赋值语句，也成了无用的废语句了。 删完之后，再次ReFormat File ### 三、基本阅读理解，加上自己理解后的注释 变成了这样： ``` def pick_high_limit(stocks, context, end_date, pre_date, pre_date_10): # 取数 df_panel = get_price(stocks, count=1, end_date=end_date, frequency='daily', fields=['open', 'close', 'high_limit', 'money', 'pre_close', 'low']) df_close = df_panel['close'] df_open = df_panel['open'] df_high_limit = df_panel['high_limit'] df_low = df_panel['low'] df_pre_close = df_panel['pre_close'] # high_limit_stock = [] # 入选的股票 for stock in stocks: if stock[0:2] == '30' or stock[0:2] == '68': # 过滤创业板和科创板 continue _high_limit = df_high_limit[stock].values _close = df_close[stock].values _open = df_open[stock].values _pre_close = df_pre_close[stock].values _low = df_low[stock].values if _close == _high_limit and _high_limit > _pre_close * 1.06: # 如果涨停且涨停的涨幅大于6%（6%显然是为了规避ST） # 5个交易日内的涨停天数 df_panel_5 = get_price(stock, count=5, end_date=end_date, frequency='daily', fields=['open', 'close', 'high_limit', 'money', 'low'], skip_paused=True) sum_plus_max_5 = (df_panel_5.loc[:, 'close'] == df_panel_5.loc[:, 'high_limit']).sum() # 前天、大前天的涨停天数 df_panel_3 = get_price(stock, count=2, end_date=pre_date, frequency='daily', fields=['open', 'close', 'high_limit', 'money', 'low'], skip_paused=True) sum_plus_max_3 = (df_panel_3.loc[:, 'close'] == df_panel_3.loc[:, 'high_limit']).sum() # 若前天、大前天的涨停天数为0，且5个交易日的涨停天数不大于2天 if sum_plus_max_3 == 0 and sum_plus_max_5 < = 2: high_limit_stock.append(stock) # 入选 return high_limit_stock ``` ### 四、理清逻辑 通过阅读理解，了解了其选股逻辑是： 选出昨日(end_date)涨停的股票，排除掉5天内涨停数大于2天的、去掉前天(pre_date)、大前天涨停过的股票。 ### 五、考虑效率 读取和处理数据，是策略程序的核心工作，这两方面也是运行效率的主要障碍，必须遵循： 1）`尽量减少读取的数据量` 1.1）减少取数的行数 - 比如其中过滤ST、创业板、科创板，应该在取数前完成，可以考虑放到基本过滤里面去； - 逐步缩小取数的范围。比如可以先排除掉昨日未涨停的股票；在缩小的范围内再去排除5天2板以上的股票；最后在进一步缩小的范围内去排除前天、大前天涨停过的股票。 1.2）减少取数的字段 - 比如第一条取数语句，取了pre_close，仅仅是为了排除ST, 取open、low，后来根本没有用上，这都是不该取的数据。 2) `能成批处理的绝不用循环去one-by-one` 一般地，应尽量消灭循环。 比如第一步，筛选出昨日涨停的股票。怎么筛选？ ```Python for stock in stocks: _high_limit = df_high_limit[stock].values _close = df_close[stock].values if _high_limit == _close: ...... ``` 这个写法，既不优雅，也不高效。换个写法： ``` # 条件1：昨天涨停 stocks = get_price(stocks, end_date=context.previous_date, fields=['close', 'high_limit', 'paused'], count=1, panel=False).query('close==high_limit and paused==0')['code'].tolist() ``` 一条语句，就搞定了。这样写，不仅仅是优雅简洁，更重要的不去存储那些中间变量，直接一步得到需要的结果，这些中间变量，也是需要占内存的。 ###六、重写代码 ``` def pick_high_limit(context, stocks): """ 选出打板的股票 """ yesterday = context.previous_date # 条件1：昨天涨停 stocks = get_price(stocks, end_date=yesterday, fields=['close', 'high_limit', 'paused'], count=1, panel=False).query('close==high_limit and paused==0')['code'].tolist() if stocks: # 条件2：昨天（含）5天内，涨停次数不超过2次 high_ups = get_price(stocks, end_date=yesterday, fields=['close', 'high_limit', 'paused'], count=5, panel=False ).query('close==high_limit and paused==0').groupby('code')['close'].count() remove_list = high_ups[high_ups > 2].index.tolist() stocks = list(set(stocks) - set(remove_list)) if stocks: # 条件3：前天，大前天，均未涨停 pre_date = get_trade_days(end_date=yesterday, count=2)[0] # 前天 high_ups = get_price(stocks, end_date=pre_date, fields=['close', 'high_limit', 'paused'], count=2, panel=False ).query('close==high_limit and paused==0').groupby('code')['close'].count() remove_list = high_ups.index.tolist() stocks = list(set(stocks) - set(remove_list)) return stocks ``` 怎么样，经过几个步骤，原来云山雾罩的程序，变得优雅简单、逻辑清晰、一目了然，是不是让人赏心悦目呢？修改后的程序运行起来，速度也是彻底秒杀原来的代码。