# Unable to call a pytensor function() from a conventional python function

**URL:** <https://discourse.pymc.io/t/unable-to-call-a-pytensor-function-from-a-conventional-python-function/13555>\
**Category:** v5\
**Tags:** pytensor\
**Created:** [January 1, 2024, 2:21pm UTC](https://discourse.pymc.io/t/unable-to-call-a-pytensor-function-from-a-conventional-python-function/13555 "2024-01-01T14:21:38Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Aravind27](https://avatars.discourse-cdn.com/v4/letter/a/c5a1d2/32.png) [@Aravind27](https://discourse.pymc.io/u/Aravind27)\
**Post date:** [January 1, 2024, 2:21pm UTC](https://discourse.pymc.io/t/unable-to-call-a-pytensor-function-from-a-conventional-python-function/13555/1 "2024-01-01T14:21:39Z")

</div>

Hello everyone,

I am using Pytensor along with Pymc for a hierarchical model. A part of my likelihood is a conditional function, so I am using Pytensor’s switch() option. I call this within a regular Python function that also handles other operations to create the likelihood function.

However, I get the following **error** when I try to do this:

```auto
File "<ipython-input-43-1cca0571e3b3>", line 1, in <cell line: 1> x = pt.vector('x') 
Expected an array-like object, but found a Variable: maybe you are trying to call a function on a (possibly shared) variable instead of a numeric array?]

```

I have made a simple version of the **code** here:

```auto
from pytensor import tensor as pt
import pymc as pm
import numpy as np

x = pt.vector('x')
t = pt.vector('t')
m1 = pt.scalar('m1')
m2 = pt.scalar('m2')
m3 = pt.scalar('m3')

y = pt.switch(pt.eq(t,5),
              x*m1,
              pt.switch(pt.eq(t,10),
                x*m2,
                x*m3)) 

y_calculator = pytensor.function([x,t,m1,m2,m3], y)

def likelihood_test(xdata,tdata,m1,m2,m3,ydata):
  mu = y_calculator(xdata, tdata, m1,m2,m3)
  return pm.Normal('y_obs', mu=mu, sigma=1, observed = ydata, shape=xdata.shape)

with pm.Model() as second_level_model:
  xdata = pm.Data('xdata', xdata_arr, mutable=True) # data
  tdata = pm.Data('tdata', tdata_arr, mutable=True) # data
  ydata = pm.Data('ydata', ydata_arr, mutable=True) # data

  m1 = pm.Lognormal('m1', mu = 1, sigma = 1) # Prior 
  # print(m1.type)
  m2 = pm.Lognormal('m2', mu = 2, sigma = 1) # Prior 
  m3 = pm.Lognormal('m3', mu = 3, sigma = 1) # Prior 
  

  likelihood_test(xdata,tdata,m1,m2,m3,ydata)

  trace = pm.sample(4000, tune = 1000, cores = 2, return_inferencedata=True)

```

Could someone please help me with this? I’m relatively new to Pytensor and still navigating my way around. Any advice on optimizing the code would also be greatly appreciated. Thank you to everyone offering their time and help here, it means a lot!

PS: I found a post that was discussing the same error. I did try to include the tensor calculations within the likelihood() function to avoid multiple compiles as recommended in the answer there. But still the same error persists.

> [@Aesara Error: Expected an array-like object, but found a Variable](https://discourse.pymc.io/t/aesara-error-expected-an-array-like-object-but-found-a-variable/10910):
>
> Hi all, I’m new to PyMC, so please excuse me if this is basic. I’ve been writing an Aesara function which calls another Aesara function. This is throwing an error and I havent found any answers in the docs or forms. Appreciate any help. Here is a simplified version of the code: x1 = at.dscalar(‘x’) y1 = at.dscalar(‘y’) z1 = x1 + y1 f1 = function([x1, y1], z1) x2 = at.dscalar(‘x’) y2 = at.dscalar(‘y’) z2 = f1(x1, x2) + 5 f2 = function([x2, y2], z2) When I compile I get this error: T…

---

<div class="post-metadata">

**Author:** ![jessegrabowski](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/jessegrabowski/32/5010_2.png) [@jessegrabowski](https://discourse.pymc.io/u/jessegrabowski)\
**Post date:** [January 1, 2024, 4:43pm UTC](https://discourse.pymc.io/t/unable-to-call-a-pytensor-function-from-a-conventional-python-function/13555/2 "2024-01-01T16:43:42Z")

</div>

You never want to compile intermediate pytensor graphs like you did with `y_calculator`. Instead, just directly use the `y` variable inside your model, like this:

```auto
with pm.Model() as second_level_model:
  xdata = pm.Data('xdata', xdata_arr, mutable=True) # data
  tdata = pm.Data('tdata', tdata_arr, mutable=True) # data
  ydata = pm.Data('ydata', ydata_arr, mutable=True) # data

  m1 = pm.Lognormal('m1', mu = 1, sigma = 1) # Prior 
  # print(m1.type)
  m2 = pm.Lognormal('m2', mu = 2, sigma = 1) # Prior 
  m3 = pm.Lognormal('m3', mu = 3, sigma = 1) # Prior 

  mu = pt.switch(pt.eq(tdata,5),
              xdata*m1,
              pt.switch(pt.eq(tdata,10),
                xdata*m2,
                xdata*m3)) 

```

If you prefer, you could make an ordinary python function that takes tensor inputs and returns tensors as outputs. So this would work:

```auto
def y_calculator(x, t, m1, m2, m3):
    return pt.switch(pt.eq(t,5),
              x*m1,
              pt.switch(pt.eq(t,10),
                x*m2,
                x*m3)) 

```

The computational graph should be complied exactly once, and only by PyMC when you call `pm.sample`

---

<div class="post-metadata">

**Author:** ![Aravind27](https://avatars.discourse-cdn.com/v4/letter/a/c5a1d2/32.png) [@Aravind27](https://discourse.pymc.io/u/Aravind27)\
**Post date:** [January 1, 2024, 6:13pm UTC](https://discourse.pymc.io/t/unable-to-call-a-pytensor-function-from-a-conventional-python-function/13555/3 "2024-01-01T18:13:27Z")

</div>

Thanks a lot @jessegrabowski. What you suggested worked flawlessly 🙂