SCI-KIT 中监督分类算法的超参数优化学习？

Question

I have NSL-KDD dataset, the task is to run various classification algorithms form scikit learn (eg: KNN-classifier), I have to get an accuracy score of more than 80% for whatever classifiers I choose, Hyperparameter optimization is to be performed, as of now if I run KNN classifier my accuracy score stands at 75.5%, what Hyperparameter Optimisation would give me an accuracy score of more than 80%??

The Data Files which are necessary to run the code:-

http://www.filedropper.com/kddtest_1

http://www.filedropper.com/kddtrain

http://www.filedropper.com/trainingattacktypes

Main Code Files:-

http://www.filedropper.com/main1_4

import os
from collections import defaultdict
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

import warnings
warnings.filterwarnings('ignore')


dataset_root = 'NSL-KDD-Dataset/NSL-KDD-Dataset'

#train_file = os.path.join(dataset_root, 'KDDTrain+.txt')
#test_file = os.path.join(dataset_root, 'KDDTest+.txt')


# Original KDD dataset feature names obtained from 
# http://kdd.ics.uci.edu/databases/kddcup99/kddcup.names
# http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html

header_names = ['duration', 'protocol_type', 'service', 'flag', 'src_bytes', 'dst_bytes', 
'land', 'wrong_fragment', 'urgent', 'hot', 'num_failed_logins', 'logged_in', 
'num_compromised', 'root_shell', 'su_attempted', 'num_root', 'num_file_creations', 
'num_shells', 'num_access_files', 'num_outbound_cmds', 'is_host_login', 'is_guest_login', 
'count', 'srv_count', 'serror_rate', 'srv_serror_rate', 'rerror_rate', 'srv_rerror_rate', 
'same_srv_rate', 'diff_srv_rate', 'srv_diff_host_rate', 'dst_host_count', 
'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate', 
'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate', 
'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate', 
'attack_type', 'success_pred']


# Differentiating between nominal, binary, and numeric features

# root_shell is marked as a continuous feature in the kddcup.names 
# file, but it is supposed to be a binary feature according to the 
# dataset documentation

col_names = np.array(header_names)

nominal_idx = [1, 2, 3]
binary_idx = [6, 11, 13, 14, 20, 21]
numeric_idx = list(set(range(41)).difference(nominal_idx).difference(binary_idx))

nominal_cols = col_names[nominal_idx].tolist()
binary_cols = col_names[binary_idx].tolist()
numeric_cols = col_names[numeric_idx].tolist()

# training_attack_types.txt maps each of the 22 different attacks to 1 of 4 categories
# file obtained from http://kdd.ics.uci.edu/databases/kddcup99/training_attack_types



category = defaultdict(list)
category['benign'].append('normal')

with open('training_attack_types.txt', 'r') as f:
    for line in f.readlines():
        attack, cat = line.strip().split(' ')
        category[cat].append(attack)

attack_mapping = dict((v,k) for k in category for v in category[k])


train_df = pd.read_csv('KDDTest+.txt', names=header_names)
train_df['attack_category'] = train_df['attack_type'] \
                                .map(lambda x: attack_mapping[x])
train_df.drop(['success_pred'], axis=1, inplace=True)

test_df = pd.read_csv('KDDTest+.txt', names=header_names)
test_df['attack_category'] = test_df['attack_type'] \
                                .map(lambda x: attack_mapping[x])
test_df.drop(['success_pred'], axis=1, inplace=True)


train_attack_types = train_df['attack_type'].value_counts()
train_attack_cats = train_df['attack_category'].value_counts()

test_attack_types = test_df['attack_type'].value_counts()
test_attack_cats = test_df['attack_category'].value_counts()

train_attack_types.plot(kind='barh', figsize=(20,10), fontsize=20)

train_attack_cats.plot(kind='barh', figsize=(20,10), fontsize=30)


test_attack_types.plot(kind='barh', figsize=(20,10), fontsize=15)

test_attack_cats.plot(kind='barh', figsize=(20,10), fontsize=30)

# Let's take a look at the binary features
# By definition, all of these features should have a min of 0.0 and a max of 1.0
#execute the commands in console

train_df[binary_cols].describe().transpose()


# Wait a minute... the su_attempted column has a max value of 2.0?

train_df.groupby(['su_attempted']).size()

# Let's fix this discrepancy and assume that su_attempted=2 -> su_attempted=0

train_df['su_attempted'].replace(2, 0, inplace=True)
test_df['su_attempted'].replace(2, 0, inplace=True)
train_df.groupby(['su_attempted']).size()


# Next, we notice that the num_outbound_cmds column only takes on one value!

train_df.groupby(['num_outbound_cmds']).size()

# Now, that's not a very useful feature - let's drop it from the dataset

train_df.drop('num_outbound_cmds', axis = 1, inplace=True)
test_df.drop('num_outbound_cmds', axis = 1, inplace=True)
numeric_cols.remove('num_outbound_cmds')


"""
Data Preparation

"""
train_Y = train_df['attack_category']
train_x_raw = train_df.drop(['attack_category','attack_type'], axis=1)
test_Y = test_df['attack_category']
test_x_raw = test_df.drop(['attack_category','attack_type'], axis=1)




combined_df_raw = pd.concat([train_x_raw, test_x_raw])
combined_df = pd.get_dummies(combined_df_raw, columns=nominal_cols, drop_first=True)

train_x = combined_df[:len(train_x_raw)]
test_x = combined_df[len(train_x_raw):]

# Store dummy variable feature names
dummy_variables = list(set(train_x)-set(combined_df_raw))

#execute the commands in console
train_x.describe()
train_x['duration'].describe()
# Experimenting with StandardScaler on the single 'duration' feature
from sklearn.preprocessing import StandardScaler

durations = train_x['duration'].values.reshape(-1, 1)
standard_scaler = StandardScaler().fit(durations)
scaled_durations = standard_scaler.transform(durations)
pd.Series(scaled_durations.flatten()).describe()

# Experimenting with MinMaxScaler on the single 'duration' feature
from sklearn.preprocessing import MinMaxScaler

min_max_scaler = MinMaxScaler().fit(durations)
min_max_scaled_durations = min_max_scaler.transform(durations)
pd.Series(min_max_scaled_durations.flatten()).describe()

# Experimenting with RobustScaler on the single 'duration' feature
from sklearn.preprocessing import RobustScaler

min_max_scaler = RobustScaler().fit(durations)
robust_scaled_durations = min_max_scaler.transform(durations)
pd.Series(robust_scaled_durations.flatten()).describe()

# Let's proceed with StandardScaler- Apply to all the numeric columns

standard_scaler = StandardScaler().fit(train_x[numeric_cols])

train_x[numeric_cols] = \
    standard_scaler.transform(train_x[numeric_cols])

test_x[numeric_cols] = \
    standard_scaler.transform(test_x[numeric_cols])

train_x.describe()


train_Y_bin = train_Y.apply(lambda x: 0 if x is 'benign' else 1)
test_Y_bin = test_Y.apply(lambda x: 0 if x is 'benign' else 1)

The KNN classifier code has to be optimised.

KNN implementation below which requires Hyperparameter optimization

from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import confusion_matrix, zero_one_loss , accuracy_score
knn_clf = KNeighborsClassifier( n_neighbors = 3 )
knn_clf.fit( train_x,train_Y )
knn_pred = knn_clf.predict(test_x)
accuracy_score(test_Y,knn_pred)

Answer 1

Well, your model may not be right for the given data set you are using. You can add more data, or tweak the parameters, like you suggested. You may want to look into overfitting or underfitting as well. As for the parameters, see my sample code below.

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import classification_report
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score
from sklearn.model_selection import GridSearchCV

df = pd.read_csv('C:\\your_path\\heart.csv')

df.head()

df.info()

df.isnull().sum()

#Univariate analysis target.
sns.countplot(df['target'])

#Univariate analysis age.
f = plt.figure(figsize=(20,4))
f.add_subplot(1,2,1)
sns.distplot(df['age'])
f.add_subplot(1,2,2)
sns.boxplot(df['age'])


#Univariate analysis resting blood pressure (mm Hg) atau trestbps.f = plt.figure(figsize=(20,4))
f.add_subplot(1,2,1)
sns.distplot(df['trestbps'])
f.add_subplot(1,2,2)
sns.boxplot(df['trestbps'])


#Create KNN Object.
from sklearn.linear_model import LogisticRegression
# all parameters not specified are set to their defaults
knn = KNeighborsClassifier()#Create x and y variables.
x = df.drop(columns=['target'])
y = df['target']#Split data into training and testing.
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=4)#Training the model.
knn.fit(x_train, y_train)#Predict test data set.

logisticRegr = LogisticRegression()
logisticRegr.fit(x_train, y_train)
# Returns a NumPy Array
# Predict for One Observation (image)
#logisticRegr.predict(x_test[0].reshape(1,-1))
logisticRegr.predict(x_test[0:10])
predictions = logisticRegr.predict(x_test)


y_pred = predictions#Checking performance our model with classification report.


print(classification_report(y_test, y_pred))#Checking performance our model with ROC Score.
roc_auc_score(y_test, y_pred)

Results:

              precision    recall  f1-score   support

           0       0.91      0.84      0.87        25
           1       0.89      0.94      0.92        36

    accuracy                           0.90        61
   macro avg       0.90      0.89      0.90        61
weighted avg       0.90      0.90      0.90        61

The performance is alright, at over 90%. But, let's try to use Hyperparameter Tuning to Improve our Model's Performance.

#List Hyperparameters that we want to tune.
leaf_size = list(range(1,50))
n_neighbors = list(range(1,30))
p=[1,2]#Convert to dictionary
hyperparameters = dict(leaf_size=leaf_size, n_neighbors=n_neighbors, p=p)#Create new KNN object
knn_2 = KNeighborsClassifier()#Use GridSearch
clf = GridSearchCV(knn_2, hyperparameters, cv=10)#Fit the model
best_model = clf.fit(x,y)#Print The value of best Hyperparameters
print('Best leaf_size:', best_model.best_estimator_.get_params()['leaf_size'])
print('Best p:', best_model.best_estimator_.get_params()['p'])
print('Best n_neighbors:', best_model.best_estimator_.get_params()['n_neighbors'])

Results:

Best leaf_size: 1
Best p: 1
Best n_neighbors: 7

Now, let's use the knowledge we accumulated above, to make a small tweak and re-run the process...

# train your model using all data and the best known parameters
# instantiate model with best parameters
knn = KNeighborsClassifier(n_neighbors=7, weights='uniform')

# fit with X and y, not X_train and y_train
# even if we use train/test split, we should train on X and y before making predictions on new data
# otherwise we throw away potential valuable data we can learn from
knn.fit(x, y)

logisticRegr = LogisticRegression()
logisticRegr.fit(x_train, y_train)
# Returns a NumPy Array
# Predict for One Observation (image)
#logisticRegr.predict(x_test[0].reshape(1,-1))
logisticRegr.predict(x_test[0:10])
predictions = logisticRegr.predict(x_test)

y_pred = predictions#Checking performance our model with classification report.

print(classification_report(y_test, y_pred))#Checking performance our model with ROC Score.
roc_auc_score(y_test, y_pred)

Results:

              precision    recall  f1-score   support

           0       0.91      0.84      0.87        25
           1       0.89      0.94      0.92        36

    accuracy                           0.90        61
   macro avg       0.90      0.89      0.90        61
weighted avg       0.90      0.90      0.90        61

It's the same thing, In this case. fiddling with the hyperparameters made no difference at all, In other cases; there may be a slight improvement in performance, 5%, 10%. or whatever, So, the takeaway is, KNN performs well on my specific data set, but apparently it doesn't deliver good results on your data set, and that's totally fine. just pick a different model to test.

# data source:
# https://raw.githubusercontent.com/adiptamartulandi/KNN-and-Tuning-Hyperparameters/master/heart.csv

I'll leave you with one final thought. You can loop through multiple classifiers automatically, and see the results of each, then pick the top 1 or 2, and run with that.

import numpy as np
import pandas as pd


# Load data from UCI dataset repo
bank_note_url = 'https://archive.ics.uci.edu/ml/machine-learning-databases/00267/data_banknote_authentication.txt'
data = np.loadtxt(bank_note_url, delimiter=',')
data = pd.DataFrame(data)

# Add column names
clean_columns = ['variance_of_wavelet', 'skewness_of_wavelet',
                 'curtosis_of_wavelet', 'entropy_of_wavelet',
                 'class']

data.columns = clean_columns

data.head()

# Separate the target and features as separate dataframes for sklearn APIs
X = data.drop('class', axis=1)
y = data[['class']].astype('int')

# Specify the design matrix and the target vector for yellowbrick as arrays
design_matrix = X.values
target_vector = y.values.flatten()

X.head()
y.head()

from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler, OneHotEncoder

numeric_transformer = Pipeline(steps=[
    ('imputer', SimpleImputer(strategy='median')),
    ('scaler', StandardScaler())])

categorical_transformer = Pipeline(steps=[
    ('imputer', SimpleImputer(strategy='constant', fill_value='missing')),
    ('onehot', OneHotEncoder(handle_unknown='ignore'))])
from sklearn.model_selection import train_test_split        

# Stratified sampling based on the distribution of the target vector, y
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    stratify=y,
                                                    test_size=0.20,
                                                    random_state=30)

numeric_features = X_train.select_dtypes(include=['int64', 'float64']).columns
categorical_features = X_train.select_dtypes(include=['object']).columns

from sklearn.compose import ColumnTransformer

preprocessor = ColumnTransformer(
    transformers=[
        ('num', numeric_transformer, numeric_features),
        ('cat', categorical_transformer, categorical_features)])


from sklearn.ensemble import RandomForestClassifier

rf = Pipeline(steps=[('preprocessor', preprocessor),
                      ('classifier', RandomForestClassifier())])

rf.fit(X_train, y_train)

y_pred = rf.predict(X_test)

from sklearn.metrics import accuracy_score, log_loss
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC, LinearSVC, NuSVC
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier, GradientBoostingClassifier
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.discriminant_analysis import QuadraticDiscriminantAnalysis

classifiers = [
    KNeighborsClassifier(3),
    SVC(kernel="rbf", C=0.025, probability=True),
    NuSVC(probability=True),
    DecisionTreeClassifier(),
    RandomForestClassifier(),
    AdaBoostClassifier(),
    GradientBoostingClassifier()
    ]

for classifier in classifiers:
    pipe = Pipeline(steps=[('preprocessor', preprocessor),
                      ('classifier', classifier)])
    pipe.fit(X_train, y_train)   
    print(classifier)
    print("model score: %.3f" % pipe.score(X_test, y_test))


param_grid = { 
    'classifier__n_estimators': [200, 500],
    'classifier__max_features': ['auto', 'sqrt', 'log2'],
    'classifier__max_depth' : [4,5,6,7,8],
    'classifier__criterion' :['gini', 'entropy']}

Result:

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=None, n_neighbors=3, p=2,
                     weights='uniform')
model score: 1.000
SVC(C=0.025, break_ties=False, cache_size=200, class_weight=None, coef0=0.0,
    decision_function_shape='ovr', degree=3, gamma='scale', kernel='rbf',
    max_iter=-1, probability=True, random_state=None, shrinking=True, tol=0.001,
    verbose=False)
model score: 0.967
NuSVC(break_ties=False, cache_size=200, class_weight=None, coef0=0.0,
      decision_function_shape='ovr', degree=3, gamma='scale', kernel='rbf',
      max_iter=-1, nu=0.5, probability=True, random_state=None, shrinking=True,
      tol=0.001, verbose=False)
model score: 0.971
DecisionTreeClassifier(ccp_alpha=0.0, class_weight=None, criterion='gini',
                       max_depth=None, max_features=None, max_leaf_nodes=None,
                       min_impurity_decrease=0.0, min_impurity_split=None,
                       min_samples_leaf=1, min_samples_split=2,
                       min_weight_fraction_leaf=0.0, presort='deprecated',
                       random_state=None, splitter='best')
model score: 0.978
C:\Users\ryans\Anaconda3\lib\site-packages\sklearn\utils\validation.py:760: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel().
  y = column_or_1d(y, warn=True)
C:\Users\ryans\Anaconda3\lib\site-packages\sklearn\utils\validation.py:760: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel().
  y = column_or_1d(y, warn=True)
C:\Users\ryans\Anaconda3\lib\site-packages\sklearn\pipeline.py:354: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples,), for example using ravel().
  self._final_estimator.fit(Xt, y, **fit_params)
RandomForestClassifier(bootstrap=True, ccp_alpha=0.0, class_weight=None,
                       criterion='gini', max_depth=None, max_features='auto',
                       max_leaf_nodes=None, max_samples=None,
                       min_impurity_decrease=0.0, min_impurity_split=None,
                       min_samples_leaf=1, min_samples_split=2,
                       min_weight_fraction_leaf=0.0, n_estimators=100,
                       n_jobs=None, oob_score=False, random_state=None,
                       verbose=0, warm_start=False)
model score: 0.993
AdaBoostClassifier(algorithm='SAMME.R', base_estimator=None, learning_rate=1.0,
                   n_estimators=50, random_state=None)
model score: 0.996
C:\Users\ryans\Anaconda3\lib\site-packages\sklearn\utils\validation.py:760: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel().
  y = column_or_1d(y, warn=True)
C:\Users\ryans\Anaconda3\lib\site-packages\sklearn\ensemble\_gb.py:1454: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel().
  y = column_or_1d(y, warn=True)
GradientBoostingClassifier(ccp_alpha=0.0, criterion='friedman_mse', init=None,
                           learning_rate=0.1, loss='deviance', max_depth=3,
                           max_features=None, max_leaf_nodes=None,
                           min_impurity_decrease=0.0, min_impurity_split=None,
                           min_samples_leaf=1, min_samples_split=2,
                           min_weight_fraction_leaf=0.0, n_estimators=100,
                           n_iter_no_change=None, presort='deprecated',
                           random_state=None, subsample=1.0, tol=0.0001,
                           validation_fraction=0.1, verbose=0,
                           warm_start=False)
model score: 0.993