important_informationML

Outlier Detection

With Quantile

• Quantile fonksiyonu ile bir feature'ın sahip olabileceği min ve max threshold belirlenir.
• Bu değer aralıklarında olmayanlar outlier kabul edilir ve bunlar verisetinden kaldırılırı.
• Böylelikle veri daha düzenli ve temiz hale gelir.

---

max_threshold=heart_data["istenilenfeature"].quantile(0.99)
max_threshold

heart_data[heart_data["istenilenfeature"]>max_threshold]

min_threshold=heart_data["istenilenfeature"].quantile(0.01)
min_threshold

df=heart_data[(heart_data["istenilenfeature"]<max_threshold) & (heart_data["istenilenfeature"]>min_threshold)]
df

---

Label Encoding

• Feature'lar string ifadeler içeriyorsa ve bunları sütun olarak kategeorik değerlere çevirmek istiyorsanız label encoding işlemini uygulayabilirsiniz.
• Örnek: iris datasetinde Species feature'ı ['Iris-setosa', 'Iris-versicolor', 'Iris-virginica']:

from sklearn.preprocessing import LabelEncoder
le=LabelEncoder()
df_le=iris.copy()
le.fit(df_le["Species"])

list(le.classes_)

df_le["Species"]=le.transform(df_le["Species"])
df_le

---

Scaling

• Decision tree için scaling yapmak şart değildir.
• Knn, k-means clustering, linear regression gibi algoritmalar scaling gerektirir.
• Scaling ile tüm değerler belli aralıklarda tutulur. Böylelikle herhangi bir değerin başka bir değeri domine etmesinin önüne geçilmiş olur.

from sklearn.preprocessing import StandardScaler
X_scale=StandardScaler().fit_transform(X)

---

Feature Selection

import matplotlib.pyplot as plt
feature_imp=pd.Series(model.feature_importances_,index=X.columns)
feature_imp.plot(kind='barh')

best_features=feature_imp.nlargest(8).index
best_features

• Feature selection, modelin eğitim süresini kısaltma avantajı sağlar.

---

Correlation | Correlation Matrix

• In the simplest terms, correlation explains to how one or more variables are related to each other.
• If there is a high correlation, it means that these two features are really related and values affect each other highly.
• If two columns show high correlation with each other, we need to remove one of them.

---

corr=final_dataset.corr()
top_corr_features=corr.index
plt.figure(figsize=(10,10))
g=sns.heatmap(final_dataset[top_corr_features].corr(),annot=True,cmap="RdYlGn")
#corr uses pearson correlation

---

Modelling

KNN ALgorithm

from sklearn.neighbors import KNeighborsClassifier
knn=KNeighborsClassifier(n_neighbors=7)
knn.fit(Xrd_train,yrd_train)
preds=knn.predict(Xrd_test)

• We should find a k which have best value for knn.

#best k for Knn

train_score=[]
test_score=[]

for i in range(1,15):
    knn=KNeighborsClassifier(n_neighbors=i)
    knn.fit(Xrd_train,yrd_train)
    
    train_score.append(knn.score(Xrd_train,yrd_train))
    test_score.append(knn.score(Xrd_test,yrd_test))

plt.figure(figsize=(12,5))
p = sns.lineplot(range(1,15),train_score,marker='*',label='Train Score')
p = sns.lineplot(range(1,15),test_score,marker='o',label='Test Score')

---

Decision Trees

• One of the many qualities of decision trees is that they require very little data preparation.
• In fact, they don't require feature scaling or centering at all.
• Decision trees are intuitive, and their decisions are easy to interpret.
• Such model are often called "white box models"
• A node's gini attribute measures its impurity: a node is "pure"(gini=0) if all training instances it applies to belong to the same class.
• Increasing min_* hyperparameters or reducing max_* hyperparameters will regularize the model.

• ---

• Veriseti en az kuralla bölünmeye çalışılır.
• Verisetinin ayrımını en iyi yapan feature en tepeye yani kök node'a yerleştirilir.
• Amaç veriyi en iyi modelleyen en küçük ağaca ulaşmaktır.

Entropy

• The concept of entropy originated in thermodynamics as a measure of molecular disorder: entropy approaches zero when molecules are still and well ordered.
• Using gini impurity or entropy, most of the time it does not make a big difference:they lead to similar trees. Gini impurity is slightly faster to compute.
• Saflık çok yüksek olursa, entropy çok düşük olur.
• Örneğin hilesiz bir madeni parada yazı veya tura gelme olasılıkları %50 ise yani eşit ise; entropy en yüksek değerinde olur.

from sklearn.tree import DecisionTreeClassifier
dt=DecisionTreeClassifier(random_state=42)

parameter={'max_leaf_nodes': range(2, 10), 'max_depth': range(1,7), 'min_samples_split' : [2,3,4]}
grid_searchDT = GridSearchCV(dt,parameter,cv=5)
grid_searchDT.fit(Xrd_train,yrd_train) 

grid_searchDT.best_params_
grid_searchDT.best_estimator_
grid_searchDT.best_score_

Visualization Tree

tree = DecisionTreeClassifier(max_depth=4, random_state=42,max_leaf_nodes=8,min_samples_split= 2)
tree.fit(Xrd_train, yrd_train)

from sklearn.tree import export_graphviz
export_graphviz(tree, out_file="tree.dot", class_names=["0", "1"],
                feature_names=None, impurity=False, filled=True)
                
import graphviz
with open("tree.dot") as f:
    dot_graph = f.read()
display(graphviz.Source(dot_graph))

---

Random Forests