Calculate GridSearchCV runtime

gridsearch_runtime.py

# runtime info based on solution below and fit_time results of the gridsearchcv return object
# based on a response on StackExchange Data Science - Naveen Vuppula
# https://datascience.stackexchange.com/a/93524/41883
# from time import time
def gridsearch_runtime(grid_obj, X_train, y_train):
  '''
  Parameters:
    grid_obj: GridSearchCV return object that has not yet been fit to training data
    X_train: split training data independent variables
    y_train: split training data containing dependent variable
  ''' 
  start = time()
  grid_obj.fit(X_train, y_train)
  end = time()
  
  mean_fit_time= grid_obj.cv_results_['mean_fit_time']
  mean_score_time= grid_obj.cv_results_['mean_score_time']
  n_splits  = grid_obj.n_splits_ #number of splits of training data
  n_iter = pd.DataFrame(grid_obj.cv_results_).shape[0] #Iterations per split
  
  time_from_cv_result = np.mean(mean_fit_time + mean_score_time) * n_splits * n_iter
  time_from_sys_time = end - start
  
  return time_from_cv_result, time_from_sys_time

gridsearch_runtime_application.py

# obviously you would use this (and write the method above) in a different manner
# but this is more for use as a very basic compute performance testing tool
# assuming some estimator (tuned_estimator), parameters, and scorer
grid_obj = GridSearchCV(tuned_estimator, parameters, scoring=acc_scorer,cv=5)

res_time, res_sys_time = gridsearch_runtime(grid_obj, X_train, y_train)

print('calculated runtime from GridSearchCV cv_results_ attributes')
print(res_time)
print('calculated runtime from using time()')
print(res_sys_time)

# Output (time units are in seconds):
#   calculated runtime from GridSearchCV cv_results_ attributes
#   293.37535548210144
#   calculated runtime from using time()
#   295.12124705314636