AttributeError: module 'SCHAModules' has no attribute 'thermodynamic'

[Tongca]

Dear developers,

When I try to start a sscha calculation or import SCHAModules I get AttributeError: module 'SCHAModules' has no attribute 'thermodynamic' The compilation seems to end well, everything is installed and I can load both cellconstructor and sscha. However, the problem is in SCHAModules as it is not seeing the thermodynamic attribute: Do you have any suggestion on how to solve it?

Thanks a lot!
Yue

[mesonepigreco]

Hi, yes we just received another user who reported a similar problem (issue #183 )
Can you give some details on how you installed the code? Was it on your local pc or on a cluster? Did you use anaconda or the builtin python version?
Thanks for your report

[Tongca]

Thank you for your reply. We installed sscha using anaconda from a cluster.
conda create -n sscha -c conda-forge python=3.10 gfortran libblas lapack openmpi julia openmpi-mpicc pip numpy scipy spglib
conda activate sscha
pip install ase julia mpi4py
pip install cellconstructor python-sscha tdscha
All compilations look great. However, I am unable to import the thermodynamic module.

Python 3.10.13 | packaged by conda-forge | (main, Dec 23 2023, 15:36:39) [GCC 12.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

import sscha
import SCHAModules
import thermodynamic
Traceback (most recent call last):
File "", line 1, in
ModuleNotFoundError: No module named 'thermodynamic'

[diegomartinez2]

I'll point out that "module thermodynamic" is in 'module_new_termodynamic.f90', so if the fortran compilation failed, there will be no module to import.
But, I do not think that's the problem because Tongca used the pip install and that sould gave an error message if the fortran compilation failed.

[Tongca]

I reinstalled sscha using anaconda in another cluster and the thermodynamics module appeared, but the problem still occurs in the current cluster.

[mesonepigreco]

It is really weird. Maybe the module is referenced incorrectly in other submodules, and this causes it to fail to be included in the shared library in some environments. Module thermodynamics is not so big and is used little in the code. Maybe we can avoid calling it in the Python code, at least the main version.

@Tongca can you report exactly the error when you try to execute the sscha?

In this way, we can help you circumvent the problem and run the calculation without using that module.

[diegomartinez2]

Hi Lorenzo, a pythonized version of that fortran code will be something like:
----------------- Thermodynamic.py ------------------------------------------

import numpy as np
from scipy.special import tanh, sinh, cosh

def w_to_a(w, T, n):
    a = np.zeros(n)
    if T == 0.0:
        a[:] = np.sqrt(1.0 / (2.0 * w))
    else:
        a[:] = np.sqrt((1.0 / tanh(0.5 * w * 315774.65221921849 / T)) / (2.0 * w))
    return a

def w_to_da(w, T, n):
    da = np.zeros(n)
    if T == 0.0:
        da[:] = -np.sqrt(1.0 / (8.0 * w**3.0))
    else:
        beta = 315774.65221921849 / T
        a = w * beta + sinh(w * beta)
        b = np.sqrt(1.0 / (32.0 * w**3.0 * (sinh(0.5 * w * beta)**3.0) * cosh(0.5 * w * beta)))
        da[:] = -a * b
    return da

def dW_f0_u0(w, T):
    if T == 0.0:
        return 0.25
    else:
        return 0.25 * (2.0 * nb(w, T) + 1.0 + 2.0 * (315774.65221921849 / T) * w * np.exp(w * 315774.65221921849 / T) * nb(w, T)**2.0)

def nb(w, T):
    if T == 0.0:
        return 0.0
    else:
        return 1.0 / (np.exp(w * 315774.65221921849 / T ) - 1.0)

---

If we put that into the python-sscha/Modules and edit the setup to take out the fortran version, it may work as expected.

EDIT: It need a little more to work... change the SSCHAmodules calls in "Ensemble.py", etc.
EDIT: sed -i 's/"SCHAModules.thermodynamic"/"Thermodynamic"/' Ensemble.py

[diegomartinez2]

The patch should be (not tested yet):

1. Create the file "Thermodynamic.py" with the code above into python-sscha/Modules

import numpy as np
from scipy.special import tanh, sinh, cosh

def w_to_a(w, T):
    n = len(w)
    a = np.zeros(n)
    if T == 0.0:
        a[:] = np.sqrt(1.0 / (2.0 * w))
    else:
        a[:] = np.sqrt((1.0 / tanh(0.5 * w * 315774.65221921849 / T)) / (2.0 * w))
    return a

2. In python-sscha/Modules run this command: sed -i 's/"SCHAModules.thermodynamic"/"Thermodynamic"/' Ensemble.py
3. Add the line import sscha.Thermodynamic as Thermodynamic at the beginning of the file Ensemble.py
   3. Optional: Edit python-sscha/setup.py to comment the old fortran code "SCHAModules/module_new_thermodynamic.f90".
4. run python setup.py install --record files.txt 2>&1 | tee install.log

[mesonepigreco]

This seems a solution. I link here also issue #183

[mesonepigreco]

@diegomartinez2 It's probably even easier, as the module thermodynamics only exports the w_to _a and w_to_da subroutines. Moreover, the only subroutine that is called from Ensemble is w_to_a, so we just need that very simple function. Moreover, we do not need the last n argument, as that is the length of w, (the python bindings automatically takes care of it).

That is, indeed, if the rest of the SCHAModules has been compiled and linked correctly against those subroutines and the only problem is in the creation of the python bindings of the module (which seems the case)

[DromaninCM]

The patch should be (not tested yet):

1. Create the file "Thermodynamic.py" with the code above into python-sscha/Modules
2. In python-sscha/Modules run this command: sed -i 's/"SCHAModules.thermodynamic"/"Thermodynamic"/' Ensemble.py
3. Optional: Edit python-sscha/setup.py to comment the old fortran code "SCHAModules/module_new_thermodynamic.f90".
4. run python setup.py install --record files.txt 2>&1 | tee install.log

Dear all,

I tried doing this but now the python-sscha is not compiling:

INFO: gfortran:f90: SCHAModules/module_anharmonic.f90
SCHAModules/module_anharmonic.f90:3.6:

use thermodynamic
1
Fatal Error: Can't open module file 'thermodynamic.mod' for reading at (1): No such file or directory

I'm attaching the install.log

Davide
install.log

[diegomartinez2]

That strange, the sed -i 's/"SCHAModules.thermodynamic"/"Thermodynamic"/' Ensemble.py must have changed the Ensemble.py. Did you do a python setup.py install --record files.txt 2>&1 | tee install.log?

In this case I will also do sed -i 's/"SCHAModules.thermodynamic"/"Thermodynamic"/' /gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/Ensemble.py to change all the SCHAModules.thermodynamic calls for Thermodynamic calls. Also put the Thermodynamic.py there /gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/Thermodynamic.py

Note that that is not the general correct way to do it, this is a quick-patch for your case.

[DromaninCM]

Dear Diego,

Even doing this doesn't work:

[romanind@ruche01 CNT_SSCHA]$ sed -i 's/"SCHAModules.thermodynamic"/"Thermodynamic"/' /gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/Ensemble.py

[romanind@ruche01 sscha]$ cd /gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/
[romanind@ruche01 sscha]$ ls
AdvancedCalculations.py Cluster.py Ensemble.py Optimizer.py Relax.py Thermodynamic.py Utilities.py pycache fourier_gradient.jl
Calculator.py Dynamical.py Minimizer.py Parallel.py SchaMinimizer.py Tools.py init.py aiida_ensemble.py

I obtain always the error:

Traceback (most recent call last):
File "/gpfs/workdir/romanind/CNT_SSCHA/hessian.py", line 48, in
dyn_hessian = ens.get_free_energy_hessian(include_v4 = INCLUDE_V4, get_full_hessian = True,verbose = False)
File "/gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/Ensemble.py", line 3657, in get_free_energy_hessian
a = SCHAModules.thermodynamic.w_to_a(w, self.current_T)
AttributeError: module 'SCHAModules' has no attribute 'thermodynamic'

I am attaching also the Ensemble.py

Davide

Ensemble.py.txt

[diegomartinez2]

Looking at your Ensemble.py file I can see that the 'sed' command didn't changed the "SCHAModules.thermodynamic" to "Thermodynamic".
Maybe you have to do it by hand? Edit the "/gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/Ensemble.py" to change that a = SCHAModules.thermodynamic.w_to_a(w, self.current_T) for a = Thermodynamic.w_to_a(w, self.current_T).

[DromaninCM]

I changed it manually:

Traceback (most recent call last):
File "/gpfs/workdir/romanind/CNT_SSCHA/hessian.py", line 48, in
dyn_hessian = ens.get_free_energy_hessian(include_v4 = INCLUDE_V4, get_full_hessian = True,verbose = False)
File "/gpfs/users/romanind/.conda/envs/Lorenzo_SSCHA/lib/python3.10/site-packages/sscha/Ensemble.py", line 3657, in get_free_energy_hessian
a = Thermodynamic.w_to_a(w, self.current_T)
NameError: name 'Thermodynamic' is not defined

I'm attaching the Ensemble.py

Davide

[Tongca]

I tried manually modifying the Ensemble.py file to change that SCHAModules.thermodynamic for Thermodynamic. In addition, “import sscha.Thermodynamic as Thermodynamic” is added to the Ensemble.py header. The error disappears when compiled.
Uploading Ensemble.txt…

[DromaninCM]

Ensemble.py.txt

[Tongca]

Have you tried adding the line "import sscha.Thermodynamic as Thermodynamic" to the beginning of the Ensemble.py?

[DromaninCM]

I tried manually modifying the Ensemble.py file to change that SCHAModules.thermodynamic for Thermodynamic. In addition, “import sscha.Thermodynamic as Thermodynamic” is added to the Ensemble.py header. The error disappears when compiled.
Uploading Ensemble.txt…

Yes, I confirm the error now disappears also in my case.
I forgot to put the header to Thermodynamic in Ensemble manually

Davide

[mesonepigreco]

Very good!
I selected Diego's answer and updated it to include all the missing steps.
We will change this in the main repository and make these changes definitive, starting from the next minor release.