Add the new feature of customized initial population

[t-harden]

What does this PR do?

Add the new feature of allowing users to specify customized initial pipeline population for TPOT.

Where should the reviewer start?

tpot/tests/test_custom_iniPop.py and tpot/tpot/base.py

How should this PR be tested?

The test code is at tpot/tests/test_custom_iniPop.py:

from tpot import TPOTClassifier
from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split

digits = load_digits()
X_train, X_test, y_train, y_test = train_test_split(digits.data, digits.target,
                                                    train_size=0.75, test_size=0.25, random_state=42)

individual_str1 = 'MultinomialNB(input_matrix, MultinomialNB__alpha=0.1, MultinomialNB__fit_prior=True)'
individual_str2 = 'GaussianNB(DecisionTreeClassifier(input_matrix, DecisionTreeClassifier__criterion=entropy, DecisionTreeClassifier__max_depth=4, DecisionTreeClassifier__min_samples_leaf=17, DecisionTreeClassifier__min_samples_split=13))'
individual_str3 = 'GaussianNB(SelectFwe(CombineDFs(input_matrix, ZeroCount(input_matrix))))'

est = TPOTClassifier(generations=3, population_size=5, verbosity=2, random_state=42, config_dict=None,
                     customized_initial_population=[individual_str1, individual_str2, individual_str3],
                      )
est.fit(X_train, y_train)
print(est.score(X_test, y_test))

You can test it by:

cd tpot
nosetests tests/test_custom_iniPop.py -s

Any background context you want to provide?

Under this version, users can specify well-defined initial pipeline population in string format by themselves. This update has the potential to enhance the algorithm's performance and reduce evolutionary time.

Several Tips:

1. These string pipelines can be obtained in two ways:

• Referencing the examples in test_custom_iniPop.py and modifying them according to TPOT's config_dict.
• Extracting the keys of self.evaluated_individuals_ evolved by TPOT. This method is particularly useful for constructing appropriate initial pipelines for better evolution.

2. We consider the relationship between #customized initial pipelines and #population as follows:

"check if #customized initial pipelines <= #population"
if len(iniPop) <= self.population_size:
    for _ in range(self.population_size - len(iniPop)):
        individual_rand = self._toolbox.individual()
        iniPop.append(individual_rand)
    print(len(customized_initial_population), "customized pipelines +", self.population_size - len(customized_initial_population), "randomized pipelines as initial population.")
else:
    raise Exception("the number of customized initial pipelines > the number of population size!")

3. We also found that in this version, the configurations (i.e., operators and parameters) of customized initial pipelines should be a subset of those specified by the config_dict parameter. This issue can be noted in the documentation or can be addressed in the near future if you agree with this PR.

What are the relevant issues?

#1321

Questions:

• Do the docs need to be updated?
  Yes
• Does this PR add new (Python) dependencies?
  No