Problem in prediction using Bayesian model in python

Question

I am using simple titanic dataset to predict the survived data using bayesian networks.Though i am able to make the structure through structure learning but after i put my test dataset after in bayesian model it shows key error as though i am able to pass the correct data in dictionary.Please refer for the bayesian model documentation: https://pgmpy.org/_modules/pgmpy/models/BayesianModel.html

from pgmpy.models import BayesianModel
from pgmpy.factors.discrete import TabularCPD, DiscreteFactor 
from pgmpy.inference import BeliefPropagation
from pgmpy.inference import VariableElimination
from pgmpy.estimators import MaximumLikelihoodEstimator,BayesianEstimator,ConstraintBasedEstimator,HillClimbSearch, BicScore,K2Score,ExhaustiveSearch


import numpy as np
import pandas as pd
import networkx as nx 
import matplotlib.pyplot as plt 
import seaborn as sns

BN_Model = BayesianModel([('Embarked', 'Fare'), ('Fare', 'Pclass'), ('Parch', 'Age'), ('Parch', 'Fare'), ('Parch', 'SibSp'), ('Parch', 'Sex'), ('Pclass', 'Survived'), ('Pclass', 'Age'), ('Sex', 'Survived'), ('SibSp', 'Fare'), ('SibSp', 'Sex')]) 
nx.draw_networkx(BN_Model,with_labels=True) 
plt.show() 

BN_Model.fit(train, estimator=MaximumLikelihoodEstimator)

test['Fare']=test['Fare'].replace(np.nan,test['Fare'].mean())

x=BN_Model.predict(test[['Embarked','Fare' ,'Parch', 'Pclass', 'Sex', 'SibSp']])```


---------------------------------------------------------------------------
_RemoteTraceback                          Traceback (most recent call last)
_RemoteTraceback: 
"""
Traceback (most recent call last):
  File "/opt/conda/lib/python3.6/site-packages/joblib/externals/loky/process_executor.py", line 418, in _process_worker
    r = call_item()
  File "/opt/conda/lib/python3.6/site-packages/joblib/externals/loky/process_executor.py", line 272, in __call__
    return self.fn(*self.args, **self.kwargs)
  File "/opt/conda/lib/python3.6/site-packages/joblib/_parallel_backends.py", line 608, in __call__
    return self.func(*args, **kwargs)
  File "/opt/conda/lib/python3.6/site-packages/joblib/parallel.py", line 256, in __call__
    for func, args, kwargs in self.items]
  File "/opt/conda/lib/python3.6/site-packages/joblib/parallel.py", line 256, in <listcomp>
    for func, args, kwargs in self.items]
  File "/opt/conda/lib/python3.6/site-packages/pgmpy/inference/ExactInference.py", line 370, in map_query
    show_progress=show_progress,
  File "/opt/conda/lib/python3.6/site-packages/pgmpy/inference/ExactInference.py", line 157, in _variable_elimination
    working_factors = self._get_working_factors(evidence)
  File "/opt/conda/lib/python3.6/site-packages/pgmpy/inference/ExactInference.py", line 44, in _get_working_factors
    [(evidence_var, evidence[evidence_var])], inplace=False
  File "/opt/conda/lib/python3.6/site-packages/pgmpy/factors/discrete/DiscreteFactor.py", line 428, in reduce
    (var, self.get_state_no(var, state_name)) for var, state_name in values
  File "/opt/conda/lib/python3.6/site-packages/pgmpy/factors/discrete/DiscreteFactor.py", line 428, in <listcomp>
    (var, self.get_state_no(var, state_name)) for var, state_name in values
  File "/opt/conda/lib/python3.6/site-packages/pgmpy/utils/state_name.py", line 74, in get_state_no
    return self.name_to_no[var][state_name]
KeyError: 7.8292
"""

The above exception was the direct cause of the following exception:

KeyError                                  Traceback (most recent call last)
<ipython-input-105-37e427dce88d> in <module>
----> 1 x=BN_Model.predict(test[['Embarked','Fare' ,'Parch', 'Pclass', 'Sex', 'SibSp']])
      2 
      3 

/opt/conda/lib/python3.6/site-packages/pgmpy/models/BayesianModel.py in predict(self, data, n_jobs)
    592             )
    593             for index, data_point in tqdm(
--> 594                 data_unique.iterrows(), total=data_unique.shape[0]
    595             )
    596         )

/opt/conda/lib/python3.6/site-packages/joblib/parallel.py in __call__(self, iterable)
   1015 
   1016             with self._backend.retrieval_context():
-> 1017                 self.retrieve()
   1018             # Make sure that we get a last message telling us we are done
   1019             elapsed_time = time.time() - self._start_time

/opt/conda/lib/python3.6/site-packages/joblib/parallel.py in retrieve(self)
    907             try:
    908                 if getattr(self._backend, 'supports_timeout', False):
--> 909                     self._output.extend(job.get(timeout=self.timeout))
    910                 else:
    911                     self._output.extend(job.get())

/opt/conda/lib/python3.6/site-packages/joblib/_parallel_backends.py in wrap_future_result(future, timeout)
    560         AsyncResults.get from multiprocessing."""
    561         try:
--> 562             return future.result(timeout=timeout)
    563         except LokyTimeoutError:
    564             raise TimeoutError()

/opt/conda/lib/python3.6/concurrent/futures/_base.py in result(self, timeout)
    430                 raise CancelledError()
    431             elif self._state == FINISHED:
--> 432                 return self.__get_result()
    433             else:
    434                 raise TimeoutError()

/opt/conda/lib/python3.6/concurrent/futures/_base.py in __get_result(self)
    382     def __get_result(self):
    383         if self._exception:
--> 384             raise self._exception
    385         else:
    386             return self._result

KeyError: 7.8292

'''

Answer 1

For prediction it is better to use the sklearn library. Although the pgmpy contains Bayesian functionalities, it serves a different goal then what your describe.

For prediction I would use following libraries:

pip install sklearn
pip install df2onehot
pip install classeval

A suggestion to make predictions:

import df2onehot
import classeval

# Import titanic dataset
X = df2onehot.import_example()
y = X['Survived']
# Remove y from X
del X['Survived']

# Make one-hot, remove numeric variables and features that contain less then 2 samples.
X = df2onehot.df2onehot(X, y_min=2)['onehot']

# Split into train test
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y)

#Import Multinomial Naive Bayes model because its all one-hot now and perhaps the most appropriate if you decide to go for Bayes.
from sklearn.naive_bayes import MultinomialNB

#Create a naive-bayes Classifier
model = MultinomialNB()

# Train the model using the training sets
model.fit(X_train, y_train)

# Predict Output
y_pred = model.predict(X_test)
y_proba = model.predict_proba(X_test)

# Evaluate results
results = classeval.eval(y_test.values.astype(bool), y_pred.astype(bool), y_proba[:,0])
classeval.plot(results)

Answer 2

Although you also describe inference, try using bnlearn for making inferences. This blog shows a step-by-step guide for structure learning and inferences.

Installation with environment:

conda create -n env_bnlearn python=3.8
conda activate env_bnlearn

pip install bnlearn

Now you can make inferences on survived like this:

import bnlearn as bn

# Load titanic dataset containing mixed variables
df_raw = bn.import_example(data='titanic')

# Pre-processing of the input dataset
dfhot, dfnum = bn.df2onehot(df_raw)

# Structure learning
DAG = bn.structure_learning.fit(dfnum)

# Plot
G = bn.plot(DAG)

# Parameter learning
model = bn.parameter_learning.fit(DAG, df)

# Print CPDs
bn.print_CPD(model)

# Make inference
q = bn.inference.fit(model, variables=['Survived'], evidence={'Sex':0, 'Pclass':1})

print(q.values)
print(q.df)

More examples can be found here.