# Usually, hyperparameter tuning is combined with cross-validation. Sometimes, we want to run cross validation independently to see whether a candidate model is generalized enough for this dataset. To this end, let's use linear regressor as an example.
#
# Code Listing 9.01. Import all the necessary packages for the cross-validation example. We use the first 150 data points of the diabetes dataset and make a linear regressor with default parameters.
from sklearn import datasets, linear_model
from sklearn.model_selection import cross_validate
diabetes = datasets.load_diabetes()
X = diabetes.data[:150]
y = diabetes.target[:150]
lr = linear_model.LinearRegression()
# Next, we will use the method cross_validate() to apply cross-validation on the linear regressor
#
# Code Listing 9.02. Use the cross_validate() method to apply 5-fold cross-validation on the linear regressor, and then print the test scores.
scores = cross_validate(lr, X, y, cv=5, scoring=('r2', 'neg_mean_squared_error'),
                       return_train_score=True)
print("negative mean squared errors: ", scores["test_neg_mean_squared_error"])
print("r2 scores: ", scores["test_r2"])
negative mean squared errors:  [-2547.29219945 -4523.25983124 -2301.49369105 -4378.07848216
 -2409.19372015]
r2 scores:  [0.36324841 0.28239194 0.4211776  0.30071196 0.61240533]
# We use 5-fold cross-validation and use r2 and negative mean square error as the metrics. As we can see from the output, the linear regressor performs differently on each fold. That's why cross-validation can help us observe the performance vibration when data changes.
#
# Now we can try to incorporate hyperparameter tuning and see how it improves the performance over the model with default parameters. We know random forest models perform very well with default settings. Can we still make improvements with hyperparameter tuning and cross-validation?
#
# Firstly, we will fetch the California housing dataset for this practice. As usual, we will randomly get 80% of the data for training.
#
# Code Listing 9.03. Fetch the California housing dataset and split it into training/test sets.
import numpy as np
from sklearn.model_selection import RandomizedSearchCV
​
from sklearn.datasets import fetch_california_housing
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, r2_score
​
california_housing_bunch = fetch_california_housing()
california_housing_X, california_housing_y = california_housing_bunch.data, california_housing_bunch.target
x_train, x_test, y_train, y_test = train_test_split(california_housing_X, california_housing_y, test_size=0.2)
# For the second step, we need to create a basic estimator. If you want to fix some hyperparameters, you can set them at this stage.
#
# train a kNN regressor with k = 10
n_neighbors=10
from sklearn.neighbors import KNeighborsRegressor
knn_10_regr = KNeighborsRegressor(n_neighbors=10)
knn_10_regr.fit(x_train, y_train)
​
# train a kNN regressor with k = 100
knn_100_regr = KNeighborsRegressor(n_neighbors=100, weights="distance")
knn_100_regr.fit(x_train, y_train)
KNeighborsRegressor(n_neighbors=100, weights='distance')
# Now it is time to create a hyperparameter grid for a random search.
#
# Code Listing 9.05. Create a hyperparameter grid for 3 parameters: n_estimators, max_depth, bootstrap for RandomForestRegressor.
#
# Number of trees in random forest
n_estimators = [int(x) for x in np.linspace(start = 600, stop = 2000, num = 15)]
# Maximum number of levels in tree
max_depth = [int(x) for x in np.linspace(10, 80, num = 8)]
max_depth.append(None)
# Method of selecting samples for training each tree
bootstrap = [True, False]
​
random_grid = {'n_neighbors': n_neighbors}
# numpy's linspace() method can create a list of numbers by pre-defined start/stop. We can start the random search + cross-validation. When we have a basic regressor, we can use it as the value for the parameter estimator for RandomizedSearchCV(), and randomly try different combinations of hyperparameters we want to test.
#
# Code Listing 9.06. Use training data to process the randomized search and cross-validation.
#
# Random search of parameters, using 3 fold cross validation,
# search across 10 different combinations, and use all available cores
rf_random = RandomizedSearchCV(estimator = knn_10_regr, param_distributions = random_grid, n_iter = 10, cv = 3, n_jobs = -1)
# Fit the random search model
rf_random.fit(x_train, y_train)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
D:\Users\psalm\AppData\Local\Temp/ipykernel_33000/966981705.py in <module>
      7 rf_random = RandomizedSearchCV(estimator = knn_10_regr, param_distributions = random_grid, n_iter = 10, cv = 3, n_jobs = -1)
      8 # Fit the random search model
----> 9 rf_random.fit(x_train, y_train)

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\utils\validation.py in inner_f(*args, **kwargs)
     61             extra_args = len(args) - len(all_args)
     62             if extra_args <= 0:
---> 63                 return f(*args, **kwargs)
     64
     65             # extra_args > 0

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\model_selection\_search.py in fit(self, X, y, groups, **fit_params)
    839                 return results
    840
--> 841             self._run_search(evaluate_candidates)
    842
    843             # multimetric is determined here because in the case of a callable

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\model_selection\_search.py in _run_search(self, evaluate_candidates)
   1631     def _run_search(self, evaluate_candidates):
   1632         """Search n_iter candidates from param_distributions"""
-> 1633         evaluate_candidates(ParameterSampler(
   1634             self.param_distributions, self.n_iter,
   1635             random_state=self.random_state))

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\utils\validation.py in inner_f(*args, **kwargs)
     61             extra_args = len(args) - len(all_args)
     62             if extra_args <= 0:
---> 63                 return f(*args, **kwargs)
     64
     65             # extra_args > 0

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\model_selection\_search.py in __init__(self, param_distributions, n_iter, random_state)
    258                 if (not isinstance(dist[key], Iterable)
    259                         and not hasattr(dist[key], 'rvs')):
--> 260                     raise TypeError('Parameter value is not iterable '
    261                                     'or distribution (key={!r}, value={!r})'
    262                                     .format(key, dist[key]))

TypeError: Parameter value is not iterable or distribution (key='n_neighbors', value=10)