Python 为什么theano.tensor.stack（）使函数编译时间快得多？

我正在写一个函数，它接受一系列酶和底物浓度，并返回一系列最大反应速率

import theano.tensor as T
from theano import function

self.rateFunction = function(enzyme_vars_array + substrate_vars_array, rateExpressionsArray, on_unused_input='ignore')

酶变量数组和底物变量数组是NOR张量的列表，使用T.dscalar（“事物的名称”）构建

RateExpressionArray是一个python表达式列表，它根据中的theano张量给出反应的动力学速率 酶变量数组和底物变量数组。因为某些酶和底物浓度可能不被使用，所以我加入了on_unused_input='ignore'标志

这些输入的大小为：

len（酶变量数组）=132

len（基板阵列）=17

len（速率表达式数组）=2402

以编写的方式运行这行代码会导致编译非常长，至少需要5分钟（我最终放弃了等待）

但是，这一更改将编译时间缩短到了几秒钟：

self.rateFunction = function(enzyme_vars_array + substrate_vars_array, T.stack(rateExpressionsArray), on_unused_input='ignore')

theano.tensor.stack（）做了什么导致编译时间发生如此巨大的变化？那么，最初的配方又是如何让它变得如此缓慢的呢

编辑：

在编写可以运行的示例代码时，这个问题变得更加奇怪。此代码生成酶变量数组、底物变量数组和rateExpressionArray的示例：

import theano.tensor as T
from theano import function
import numpy as np


enzyme_names = ['e1','e2','e3','e4','e5','e6','e7','e8','e9','e10','e11','e12','e13','e14','e15','e16','e17','e18','e19','e20', '21','e22','e23','e24','e25','e26','e27','e28','e29','e30','e31','e32','e33','e34','e35','e36','e37','e38','e39','e40','e41','e42','e43','e44','e45','e46','e47','e48','e49','e50','e61','e62','e63','e64','e65','e66','e67','e68','e69','e70','e71','e72','e73','e74','e75','e76','e77','e78','e79','e80','e81','e82','e83','e84','e85','e86','e87','e88','e89','e90','e91','e92','e93','e94','e95','e96','e97','e98','e99','e100','e101','e102','e103','e104','e105','e106','e107','e108','e109','e110','e111','e112','e113','e114','e115','e116','e117','e118','e119','e120','e121','e122','e123','e124','e125','e126','e127','e128','e129','e130','e131','e132']

substrate_names = ['s1','s2','s3','s4','s5','s6','s7','s8','s9','s10','s11','s12','s13','s14','s15','s16','s17']

enzyme_vars_array = T.dscalars(*enzyme_names)

substrate_vars_array = T.dscalars(*substrate_names)


rateExpressionsArray = [0]*2292

# Populate rateExpressionsArray with random expressions in terms of the theano variables
for index, rateExpression in enumerate(rateExpressionsArray):
    random_enzyme_index = np.random.randint(len(enzyme_vars_array))
    random_substrate_index = np.random.randint(len(substrate_vars_array))

    rateExpressionsArray[index] = enzyme_vars_array[random_enzyme_index] * substrate_vars_array[random_substrate_index]


rateFunction = function(enzyme_vars_array + substrate_vars_array, rateExpressionsArray, on_unused_input='ignore')

这是我希望缓慢运行的代码（因为它在没有T.stack的情况下运行）。然而，在本例中，它现在以完全合理的速度运行。但是，当我将最后一行替换为以下内容时：

rateFunction = function(enzyme_vars_array + substrate_vars_array, T.stack(rateExpressionsArray), on_unused_input='ignore')

它再次运行得非常慢，与我在实际代码中观察到的行为相反

我考虑了这个示例与实际代码之间可能存在的差异，并意识到我的real RateExpressionArray非常稀疏，它主要由一个占位符组成，该占位符只返回一个默认值，只有几个实际表达式。因此，重写我的示例代码，使其看起来更像这样，我有：

import theano.tensor as T
from theano import function
import numpy as np


enzyme_names = ['e1','e2','e3','e4','e5','e6','e7','e8','e9','e10','e11','e12','e13','e14','e15','e16','e17','e18','e19','e20', '21','e22','e23','e24','e25','e26','e27','e28','e29','e30','e31','e32','e33','e34','e35','e36','e37','e38','e39','e40','e41','e42','e43','e44','e45','e46','e47','e48','e49','e50','e61','e62','e63','e64','e65','e66','e67','e68','e69','e70','e71','e72','e73','e74','e75','e76','e77','e78','e79','e80','e81','e82','e83','e84','e85','e86','e87','e88','e89','e90','e91','e92','e93','e94','e95','e96','e97','e98','e99','e100','e101','e102','e103','e104','e105','e106','e107','e108','e109','e110','e111','e112','e113','e114','e115','e116','e117','e118','e119','e120','e121','e122','e123','e124','e125','e126','e127','e128','e129','e130','e131','e132']

substrate_names = ['s1','s2','s3','s4','s5','s6','s7','s8','s9','s10','s11','s12','s13','s14','s15','s16','s17']

enzyme_vars_array = T.dscalars(*enzyme_names)

substrate_vars_array = T.dscalars(*substrate_names)


# This is the placeholder expression which returns an input unmodified
noRate = T.dscalar('noRate')
rateExpressionsArray = [noRate]*2292

# Include only a handful of nontrivial expressions
for index in xrange(1,50):
    random_enzyme_index = np.random.randint(len(enzyme_vars_array))
    random_substrate_index = np.random.randint(len(substrate_vars_array))
    rateExpressionsArray[index] = enzyme_vars_array[random_enzyme_index] * substrate_vars_array[random_substrate_index]


rateFunction = function(enzyme_vars_array + substrate_vars_array + [noRate], T.stack(rateExpressionsArray), on_unused_input='ignore')

这产生了原始行为—当rateExpressionArray中只有少数索引是非平凡表达式时，代码使用T.stack（）运行得很快，而不使用T.stack（）运行得很慢。要复制慢速版本，请将最后一行替换为：

rateFunction = function(enzyme_vars_array + substrate_vars_array + [noRate], rateExpressionsArray, on_unused_input='ignore')

---

看来T.stack在稀疏输入和密集输入上产生了不同的行为？现在我对stack（）的函数和机制更加困惑。如果您能了解这里发生的一切，我们将不胜感激

你能分享更多你的代码吗？或者，也许更好的是，您能找到一个更简单的示例，可以复制、粘贴和运行，而无需依赖项，以演示该行为吗？您能分享更多的代码吗？或者，也许更好的是，您能找到一个更简单的示例来演示该行为，该示例可以在没有依赖项的情况下复制、粘贴和运行？