model.load() IndexError: list index out of range

[ZmuX1n]

After I save the trained model and execute model.load(), it fails with "list index out of range". Here is my code. Can you tell me what's the problem?
model.train([data_df])
model.save("model")
model = model.load("model")

[yangwenz]

Hi, could you provide more information about this problem, e.g., what is "model"?

[ZmuX1n]

Thanks for your reply. Here is my complete code:
model = BayesianNetwork(config=BayesianNetwork.config_class(graph=graph_df))
model.train([data_df])
model.save("model")
model = model.load("model")
Then, it fails with the IndexError above.
Besides, I tried for "model = BayesianNetwork.load("model")", but failed with the same error

[ZmuX1n]

If I don't use save-->load function, like:
model = BayesianNetwork(config=BayesianNetwork.config_class(graph=graph_df))
model.train([data_df])
model.find_root_cause(["..."])
It performs well. So I think there should be something wrong with the "load" function

[yangwenz]

Hi, could you copy all the error messages here? It will help to locate the problem.

[ZmuX1n]

Ok, here are all the error messages:
IndexError Traceback (most recent call last)
Cell In[7], line 1
----> 1 model = BayesianNetwork.load("model")

File D:\Anaconda3\envs\python38\lib\site-packages\pyrca\analyzers\bayesian.py:442, in BayesianNetwork.load(cls, directory, filename, **kwargs)
439 @classmethod
440 def load(cls, directory, filename="bn", **kwargs):
441 model = cls.new(cls)
--> 442 reader = BIFReader(os.path.join(directory, f"{filename}.bif"))
443 model.bayesian_model = reader.get_model()
444 with open(os.path.join(directory, f"{filename}_info.pkl"), "rb") as f:

File D:\Anaconda3\envs\python38\lib\site-packages\pyrca\thirdparty\pgmpy\readwrite\BIF.py:90, in BIFReader.init(self, path, string, include_properties, n_jobs)
88 self.probability_expr, self.cpd_expr = self.get_probability_grammar()
89 self.network_name = self.get_network_name()
---> 90 self.variable_names = self.get_variables()
91 self.variable_states = self.get_states()
92 if self.include_properties:

File D:\Anaconda3\envs\python38\lib\site-packages\pyrca\thirdparty\pgmpy\readwrite\BIF.py:200, in BIFReader.get_variables(self)
198 variable_names = []
199 for block in self.variable_block():
--> 200 name = self.name_expr.searchString(block)[0][0]
201 variable_names.append(name)
203 return variable_names

File D:\Anaconda3\envs\python38\lib\site-packages\pyparsing\results.py:193, in ParseResults.getitem(self, i)
191 def getitem(self, i):
192 if isinstance(i, (int, slice)):
--> 193 return self._toklist[i]
194 else:
195 if i not in self._all_names:

IndexError: list index out of range

[yangwenz]

The problem comes from BIFReader. Is it OK to share the dumped .bif file so that we can examine this issue?

[ZmuX1n]

I'm glad to share the .bif file.
Should I send the file to your email or by some other methods?

[yangwenz]

Thanks! You can send me email ([email protected]) or share it via Dropbox or Google drive.

[ZmuX1n]

here is the google drive link: https://drive.google.com/file/d/1VPbhif8NJP3-1Y5ZkK5XonXPrYgNTbIA/view?usp=sharing

[yangwenz]

The problem comes from the naming issue. The metric names must be English or utf-8 encoded. Otherwise, the parser of pgmpy will fail.

[ZmuX1n]

I got it. So, how can I resolve the problem if my dataset must be Chinese?

[yangwenz]

You can rename the dataframe columns as x1, x2, ..., and make a mapping from chinese names to these names.

[ZmuX1n]

It's a good idea! Thanks for your guidance. I will try this method, and I am looking forward to the follow-up communication with you!