WorldQuant101因子库
有前两天的函数积累，我们直接添加因子即可：
大家仔细看会发现，WorldQuant101的因子，思路上“雷同”的居多。
就是价量关系，排序后取相关性，也就是“价量背离”。
names.append('alpha012')
features.append('(sign(delta(volume, 1)) * (-1 * delta(close, 1)))')
names.append('alpha013')
features.append('(-1 * rank(covariance(rank(close), rank(volume), 5))) ')
names.append('alpha014')
features.append('((-1 * rank(delta(returns, 3))) * correlation(open, volume, 10))')
names.append('alpha015')
features.append('(-1 * sum(rank(correlation(rank(high), rank(volume), 3)), 3))')
比如，29和30号因子，这么长：
Alpha#29: (min(product(rank(rank(scale(log(sum(ts_min(rank(rank((-1 * rank(delta((close - 1), 5))))), 2), 1))))), 1), 5) + ts_rank(delay((-1 * returns), 6), 5))
Alpha#30: (((1.0 - rank(((sign((close - delay(close, 1))) + sign((delay(close, 1) - delay(close, 2)))) + sign((delay(close, 2) - delay(close, 3)))))) * sum(volume, 5)) / sum(volume, 20))
咱们的表达式引擎计算起来也毫无压力。
第31号因子，涉及到一个新函数 decay_linear，我们需要实现即可：
Alpha#31: ((rank(rank(rank(decay_linear((-1 * rank(rank(delta(close, 10)))), 10)))) + rank((-1 * delta(close, 3)))) + sign(scale(correlation(adv20, low, 12))))
把WorldQuant 101刷了一轮，补充了所有的运算函数：
from datafeed.expr_functions import * class AlphaBase: pass # https://www.joinquant.com/data/dict/alpha101 class WorldQuant101(AlphaBase): def get_names_features(self): names = [] features = [] # names.append('alpha001') # features.append('(rank(ts_argmax(signed_power((stddev(returns, 20) if (returns < 0) else close), 2.), ' # '5)) - 0.5)') names.append('alpha002') features.append('(-1 * correlation(rank(delta(log(volume), 2)), rank(((close - open) / open)), 6))') names.append('alpha003') features.append('(-1 * correlation(rank(open), rank(volume), 10))') names.append('alpha004') features.append('(-1 * ts_rank(rank(low), 9))') ''' Alpha#5: (rank((open - (sum(vwap, 10) / 10))) * (-1 * abs(rank((close - vwap))))) Alpha#6: (-1 * correlation(open, volume, 10)) ''' names.append('alpha006') features.append('(-1 * correlation(open, volume, 10))') ''' Alpha#7: ((adv20 < volume) ? ((-1 * ts_rank(abs(delta(close, 7)), 60)) * sign(delta(close, 7))) : (-1 * 1)) Alpha#8: (-1 * rank(((sum(open, 5) * sum(returns, 5)) - delay((sum(open, 5) * sum(returns, 5)), 10)))) ''' names.append('alpha008') features.append( '(-1 * rank(((sum(open, 5) * sum(returns, 5)) - delay((sum(open, 5) * sum(returns, 5)), 10))))') ''' Alpha#11: ((rank(ts_max((vwap - close), 3)) + rank(ts_min((vwap - close), 3))) * rank(delta(volume, 3))) Alpha#12: (sign(delta(volume, 1)) * (-1 * delta(close, 1))) Alpha#13: (-1 * rank(covariance(rank(close), rank(volume), 5))) Alpha#14: ((-1 * rank(delta(returns, 3))) * correlation(open, volume, 10)) Alpha#15: (-1 * sum(rank(correlation(rank(high), rank(volume), 3)), 3)) ''' names.append('alpha012') features.append('(sign(delta(volume, 1)) * (-1 * delta(close, 1)))') names.append('alpha013') features.append('(-1 * rank(covariance(rank(close), rank(volume), 5))) ') names.append('alpha014') features.append('((-1 * rank(delta(returns, 3))) * correlation(open, volume, 10))') names.append('alpha015') features.append('(-1 * sum(rank(correlation(rank(high), rank(volume), 3)), 3))') ''' Alpha#18: (-1 * rank(((stddev(abs((close - open)), 5) + (close - open)) + correlation(close, open, 10)))) Alpha#19: ((-1 * sign(((close - delay(close, 7)) + delta(close, 7)))) * (1 + rank((1 + sum(returns, 250))))) Alpha#20: (((-1 * rank((open - delay(high, 1)))) * rank((open - delay(close, 1)))) * rank((open - delay(low, 1)))) ''' names.append('alpha018') features.append('(-1 * rank(((stddev(abs((close - open)), 5) + (close - open)) + correlation(close, open, 10))))') names.append('alpha019') features.append( '((-1 * sign(((close - delay(close, 7)) + delta(close, 7)))) * (1 + rank((1 + sum(returns, 250)))))') names.append('alpha020') features.append( '(((-1 * rank((open - delay(high, 1)))) * rank((open - delay(close, 1)))) * rank((open - delay(low, 1))))') ''' Alpha#22: (-1 * (delta(correlation(high, volume, 5), 5) * rank(stddev(close, 20)))) Alpha#26: (-1 * ts_max(correlation(ts_rank(volume, 5), ts_rank(high, 5), 5), 3)) Alpha#28: scale(((correlation(adv20, low, 5) + ((high + low) / 2)) - close)) Alpha#29: (min(product(rank(rank(scale(log(sum(ts_min(rank(rank((-1 * rank(delta((close - 1), 5))))), 2), 1))))), 1), 5) + ts_rank(delay((-1 * returns), 6), 5)) Alpha#30: (((1.0 - rank(((sign((close - delay(close, 1))) + sign((delay(close, 1) - delay(close, 2)))) + sign((delay(close, 2) - delay(close, 3)))))) * sum(volume, 5)) / sum(volume, 20)) ''' names.append('alpha022') features.append( '(-1 * (delta(correlation(high, volume, 5), 5) * rank(stddev(close, 20))))') names.append('alpha026') features.append( '(-1 * ts_max(correlation(ts_rank(volume, 5), ts_rank(high, 5), 5), 3))') #names.append('alpha028') #features.append( # 'scale(((correlation(adv20, low, 5) + ((high + low) / 2)) - close))') names.append('alpha029') features.append( '(min(product(rank(rank(scale(log(sum(ts_min(rank(rank((-1 * rank(delta((close - 1), 5))))), 2), 1))))), 1), 5) + ts_rank(delay((-1 * returns), 6), 5))') names.append('alpha030') features.append('(((1.0 - rank(((sign((close - delay(close, 1))) + sign((delay(close, 1) - delay(close, 2)))) + sign((delay(close, 2) - delay(close, 3)))))) * sum(volume, 5)) / sum(volume, 20))') ''' Alpha#31: ((rank(rank(rank(decay_linear((-1 * rank(rank(delta(close, 10)))), 10)))) + rank((-1 * delta(close, 3)))) + sign(scale(correlation(adv20, low, 12)))) Alpha#32: (scale(((sum(close, 7) / 7) - close)) + (20 * scale(correlation(vwap, delay(close, 5), 230)))) Alpha#33: rank((-1 * ((1 - (open / close))^1))) Alpha#34: rank(((1 - rank((stddev(returns, 2) / stddev(returns, 5)))) + (1 - rank(delta(close, 1))))) Alpha#35: ((Ts_Rank(volume, 32) * (1 - Ts_Rank(((close + high) - low), 16))) * (1 - Ts_Rank(returns, 32))) ''' names.append('alpha034') features.append( 'rank(((1 - rank((stddev(returns, 2) / stddev(returns, 5)))) + (1 - rank(delta(close, 1)))))') names.append('alpha035') features.append( '((ts_rank(volume, 32) * (1 - ts_rank(((close + high) - low), 16))) * (1 - ts_rank(returns, 32)))') ''' Alpha#36: (((((2.21 * rank(correlation((close - open), delay(volume, 1), 15))) + (0.7 * rank((open - close)))) + (0.73 * rank(Ts_Rank(delay((-1 * returns), 6), 5)))) + rank(abs(correlation(vwap, adv20, 6)))) + (0.6 * rank((((sum(close, 200) / 200) - open) * (close - open))))) Alpha#37: (rank(correlation(delay((open - close), 1), close, 200)) + rank((open - close))) Alpha#38: ((-1 * rank(Ts_Rank(close, 10))) * rank((close / open))) Alpha#39: ((-1 * rank((delta(close, 7) * (1 - rank(decay_linear((volume / adv20), 9)))))) * (1 + rank(sum(returns, 250)))) Alpha#40: ((-1 * rank(stddev(high, 10))) * correlation(high, volume, 10)) ''' names.append('alpha037') features.append( '(rank(correlation(delay((open - close), 1), close, 200)) + rank((open - close)))') names.append('alpha038') features.append( '((-1 * rank(ts_rank(close, 10))) * rank((close / open)))') names.append('alpha040') features.append( '((-1 * rank(stddev(high, 10))) * correlation(high, volume, 10))') ''' Alpha#44: (-1 * correlation(high, rank(volume), 5)) Alpha#45: (-1 * ((rank((sum(delay(close, 5), 20) / 20)) * correlation(close, volume, 2)) * rank(correlation(sum(close, 5), sum(close, 20), 2)))) ''' names.append('alpha044') features.append( '(-1 * correlation(high, rank(volume), 5))') names.append('alpha045') features.append( '(-1 * ((rank((sum(delay(close, 5), 20) / 20)) * correlation(close, volume, 2)) * rank(correlation(sum(close, 5), sum(close, 20), 2))))') ''' Alpha#52: ((((-1 * ts_min(low, 5)) + delay(ts_min(low, 5), 5)) * rank(((sum(returns, 240) - sum(returns, 20)) / 220))) * ts_rank(volume, 5)) Alpha#53: (-1 * delta((((close - low) - (high - close)) / (close - low)), 9)) Alpha#54: ((-1 * ((low - close) * (open^5))) / ((low - high) * (close^5))) Alpha#55: (-1 * correlation(rank(((close - ts_min(low, 12)) / (ts_max(high, 12) - ts_min(low, 12)))), rank(volume), 6)) Alpha#60: (0 - (1 * ((2 * scale(rank(((((close - low) - (high - close)) / (high - low)) * volume)))) - scale(rank(ts_argmax(close, 10)))))) ''' names.append('alpha052') features.append( '((((-1 * ts_min(low, 5)) + delay(ts_min(low, 5), 5)) * rank(((sum(returns, 240) - sum(returns, 20)) / 220))) * ts_rank(volume, 5))') names.append('alpha053') features.append( '(-1 * delta((((close - low) - (high - close)) / (close - low)), 9))') names.append('alpha055') features.append( '(-1 * correlation(rank(((close - ts_min(low, 12)) / (ts_max(high, 12) - ts_min(low, 12)))), rank(volume), 6))') names.append('alpha060') features.append('(0 - (1 * ((2 * scale(rank(((((close - low) - (high - close)) / (high - low)) * volume)))) - scale(rank(ts_argmax(close, 10))))))') names.append('alpha101') features.append('((close - open) / ((high - low) + .001))') return names, features
几点收获：
一、积累运算函数库，后续我们机器挖掘可以复用，丢给gplearn和强化学习。
二、因子构造的思路，其实这个我之前让GPT做过：
今天文章涉及的代码在如下位置：
量化数据中心建设
昨天有同学问数据的事情，咱们公众号对应的文章，使用到的数据，都会随代码一起打包在星球发布。
但为了更方便使用，咱们可以考虑建设一个会自动更新的数据中台。
这也印证了咱们是想把AI量化当作事业持续做下去的。
大模型LLM
最近大模型领域，重磅消息是Lalla3开源。
目前看来，llama3-8B的版本，很多小团队有几张卡就可以玩，这是个令人振奋的消息。
可以做的事情有三件： 增量预训练（比如加入更多的中文数据集或者领域数据集）_如果只是小数据集，在增量预训练中如何起作用？指令微调（让它更加能听懂人话），SFT(监督微调）。
需要自己构建数据集，指令集，领域数据集等，同时还需要有评测手段，以评估训练的结果。
另外，就是基于大模型的api，使用指令（prompts）与大模型交互。这个优点是自己要做的事情比较少，调API就好了，缺点是模型不符合预期时，没法干预，还是token还是比较贵的。
吾日三省吾身
最近老看到一些职场里，竞业限制引发的争端，甚至很多人赔了不少钱。
以前这种事情可能只会发生于大公司，高管圈之中。而且多数这些钱对他们来讲，小case。更多是为了专利保护，技术保密之类的诉求。
现在似乎情况发生了变化。
有些人就是普通的开发，运营。
以前的互联网，人员流动比较快。甚至很多公司在做类似项目时，就高价到对方公司挖人。
以前的互联网，开除人也比较快。现在随着新劳动法的出台，基本上没有N+1是结束不了。
企业无法高效激发活力，这是一种负担。人才不能有效流动。这也是一种负担。
不过从成熟的，契约精神的角度，平衡工作与生活的角度，也许未必是错的。
毕竟，工作都是为了更好地生活嘛。
发布者：股市刺客，转载请注明出处：https://www.95sca.cn/archives/103409
站内所有文章皆来自网络转载或读者投稿，请勿用于商业用途。如有侵权、不妥之处，请联系站长并出示版权证明以便删除。敬请谅解！
