Follow up to issue #892

Hello -

This is in reference to [#892 ]

When run standard TPOT code with `warm_start=False` for say `generations=3` and then repeat my demo (see #892) using `'warm_start=True`, I am getting different numbers of `evaluated_individuals_`. Here is the quick repeated code

`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)`

`pipeline_optimizer = TPOTClassifier(generations=3, population_size=5,
random_state=42, verbosity=2, warm_start=False)`
`pipeline_optimizer.fit(X_train, y_train)`
`total_evaluated = len(pipeline_optimizer.evaluated_individuals_) # 18`

Comparing -

`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) `

`pipeline_optimizer = TPOTClassifier(generations=1, population_size=5,
random_state=42, verbosity=2, warm_start=True)`
`total_evaluated = 0`
`for i in range(3): # where 3 is a proxy for gens`
`pipeline_optimizer.fit(X_train, y_train)`
`total_evaluated = total_evaluated + len(pipeline_optimizer.evaluated_individuals_)`
`total_evaluated # 27 (9*3)`

Am I missing something here or am I double/triple counting in my demo code?

Also is my understanding of evaluated_individuals_ wrong? Shouldn't the total number of pipelines (population_size + generations × offspring_size ) match the length of evaluated_individuals_? Here offspring_size is replaced by population_size by default. I would expect that total pipelines evaluated = 5 + (3x5) = 20 which by the way is indicated in the progress bar `_pbar` and not 18

Thanks so much in advanced for providing the support, answers and everything you do for the community !