Train set for anomaly detection with PCA about handson-unsupervised-learning HOT 4 CLOSED

aapatel09 commented on August 17, 2024

Train set for anomaly detection with PCA

from handson-unsupervised-learning.

Comments (4)

aapatel09 commented on August 17, 2024 1

Thanks. If you remove the random_state and run the experiment a few times, how much do the average precisions vary? Do the results improve as you adjust the number of principal components? If you do not split the train and test in the same way as the book, the results will vary. But, using the levers above, you should be able to find a more optimal solution. Let me know, and I could help from there. Ankur

…

On Thu, Mar 5, 2020 at 8:39 AM Steve Azzolin ***@***.***> wrote: I am trying to do anomaly detection with Normal PCA and 27 principal components as presented in the book but if I run the model on the entire dataset (without splitting into train and set) I get a completely different result: data = pd.read_csv('credit_card.zip') X = data.copy().drop(["Class"],axis=1) y = data['Class'].copy() featuresToScale = X.columns scaler = StandardScaler(copy=True) X.loc[:,featuresToScale] = scaler.fit_transform(X[featuresToScale]) pca = PCA(n_components=n_components, random_state=42) X_train_PCA = pca.fit_transform(X) X_train_PCA = pd.DataFrame(data=X_train_PCA, index=X.index) X_train_PCA_inverse = pca.inverse_transform(X_train_PCA) X_train_PCA_inverse = pd.DataFrame(data=X_train_PCA_inverse, index=X.index) anomalyScoresPCA = anomalyScores(X, X_train_PCA_inverse) plotResults(y, anomalyScoresPCA, False) [image: plot produced by the code] <https://user-images.githubusercontent.com/53196079/75986458-66507e00-5eee-11ea-92d7-784fd4d94cc3.png> Do you have any ideas on why this appens ? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#10?email_source=notifications&email_token=AEDDVJAYHV32QEY4JDN44KTRF6TS5A5CNFSM4LCKDU62YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4ISZBQMQ>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEDDVJA572RGAEREDXLMCULRF6TS5ANCNFSM4LCKDU6Q> .

from handson-unsupervised-learning.

aapatel09 commented on August 17, 2024 1

Got it. Could you try running this on just the train set per the book? I want to make sure it's the data split that's causing the discrepancy, rather than something else. Let me know!

…

On Thu, Mar 5, 2020 at 9:26 AM Steve Azzolin ***@***.***> wrote: I have manually executed the code without random_state and the score doesn't change actually (keeping n_components to 27). Then I tried to change n_component: import numpy as np import pandas as pd from sklearn.preprocessing import StandardScaler from sklearn.decomposition import PCA from sklearn.metrics import precision_recall_curve, average_precision_score from sklearn.model_selection import train_test_split import matplotlib.pyplot as plt %matplotlib inline data = pd.read_csv('credit_card.zip') X = data.copy().drop(["Class"],axis=1) y = data['Class'].copy() featuresToScale = X.columns scaler = StandardScaler(copy=True) X.loc[:,featuresToScale] = scaler.fit_transform(X[featuresToScale]) for i in [i+10 for i in range(20)]: pca = PCA(n_components=i) X_train_PCA = pca.fit_transform(X) X_train_PCA = pd.DataFrame(data=X_train_PCA, index=X.index) X_train_PCA_inverse = pca.inverse_transform(X_train_PCA) X_train_PCA_inverse = pd.DataFrame(data=X_train_PCA_inverse, index=X.index) anomalyScoresPCA = anomalyScores(X, X_train_PCA_inverse) _ , avg_precision = plotResults(y, anomalyScoresPCA, True,False) print("n_components=" , i , " avg_precision=",avg_precision) Gives: n_components= 10 avg_precision= 0.15052200650486952 n_components= 11 avg_precision= 0.1851608802374098 n_components= 12 avg_precision= 0.1820248810643287 n_components= 13 avg_precision= 0.12039760005966599 n_components= 14 avg_precision= 0.13113627677672549 n_components= 15 avg_precision= 0.22555564098028213 n_components= 16 avg_precision= 0.21592680643385742 n_components= 17 avg_precision= 0.10970330227982492 n_components= 18 avg_precision= 0.1420103680819224 n_components= 19 avg_precision= 0.11227862548711573 n_components= 20 avg_precision= 0.15777373139629802 n_components= 21 avg_precision= 0.1592625134410127 n_components= 22 avg_precision= 0.15900773562327336 n_components= 23 avg_precision= 0.19549141342644125 n_components= 24 avg_precision= 0.10344835746696714 n_components= 25 avg_precision= 0.09745140764930849 n_components= 26 avg_precision= 0.11459766332143032 n_components= 27 avg_precision= 0.30317062682018747 n_components= 28 avg_precision= 0.002776650831794116 n_components= 29 avg_precision= 0.009325957416360619 — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#10?email_source=notifications&email_token=AEDDVJGHZOXOSJXSKTD46X3RF6Y7ZA5CNFSM4LCKDU62YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEN5OFDQ#issuecomment-595255950>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEDDVJHLKOU7C5X26BXS4SDRF6Y7ZANCNFSM4LCKDU6Q> .

from handson-unsupervised-learning.

steveazzolin commented on August 17, 2024

I have manually executed the code without random_state and the score doesn't change actually (keeping n_components to 27).

Then I tried to change n_component:

import numpy as np
import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.metrics import precision_recall_curve, average_precision_score
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
%matplotlib inline

data = pd.read_csv('credit_card.zip')
X = data.copy().drop(["Class"],axis=1)
y = data['Class'].copy()

featuresToScale = X.columns
scaler = StandardScaler(copy=True)
X.loc[:,featuresToScale] = scaler.fit_transform(X[featuresToScale])

for i in [i+10 for i in range(20)]:
    pca = PCA(n_components=i)

    X_train_PCA = pca.fit_transform(X)
    X_train_PCA = pd.DataFrame(data=X_train_PCA, index=X.index)

    X_train_PCA_inverse = pca.inverse_transform(X_train_PCA)
    X_train_PCA_inverse = pd.DataFrame(data=X_train_PCA_inverse, index=X.index)

    anomalyScoresPCA = anomalyScores(X, X_train_PCA_inverse)
    _ , avg_precision = plotResults(y, anomalyScoresPCA, True,False) 
    print("n_components=" , i , " avg_precision=",avg_precision)

Gives:

n_components= 10  avg_precision= 0.15052200650486952
n_components= 11  avg_precision= 0.1851608802374098
n_components= 12  avg_precision= 0.1820248810643287
n_components= 13  avg_precision= 0.12039760005966599
n_components= 14  avg_precision= 0.13113627677672549
n_components= 15  avg_precision= 0.22555564098028213
n_components= 16  avg_precision= 0.21592680643385742
n_components= 17  avg_precision= 0.10970330227982492
n_components= 18  avg_precision= 0.1420103680819224
n_components= 19  avg_precision= 0.11227862548711573
n_components= 20  avg_precision= 0.15777373139629802
n_components= 21  avg_precision= 0.1592625134410127
n_components= 22  avg_precision= 0.15900773562327336
n_components= 23  avg_precision= 0.19549141342644125
n_components= 24  avg_precision= 0.10344835746696714
n_components= 25  avg_precision= 0.09745140764930849
n_components= 26  avg_precision= 0.11459766332143032
n_components= 27  avg_precision= 0.30317062682018747
n_components= 28  avg_precision= 0.002776650831794116
n_components= 29  avg_precision= 0.009325957416360619

from handson-unsupervised-learning.

steveazzolin commented on August 17, 2024

Running this code

import numpy as np
import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.metrics import precision_recall_curve, average_precision_score
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
%matplotlib inline

data = pd.read_csv('credit_card.zip')
X = data.copy().drop(["Class"],axis=1)
y = data['Class'].copy()

featuresToScale = X.columns
scaler = StandardScaler(copy=True)
X.loc[:,featuresToScale] = scaler.fit_transform(X[featuresToScale])

X, X_test, y, y_test = train_test_split(X, y, test_size=0.33, random_state=2018, stratify=y)

for i in [i+10 for i in range(20)]:
    pca = PCA(n_components=i,random_state=2018)

    X_train_PCA = pca.fit_transform(X)
    X_train_PCA = pd.DataFrame(data=X_train_PCA, index=X.index)

    X_train_PCA_inverse = pca.inverse_transform(X_train_PCA)
    X_train_PCA_inverse = pd.DataFrame(data=X_train_PCA_inverse, index=X.index)

    anomalyScoresPCA = anomalyScores(X, X_train_PCA_inverse)
    _ , avg_precision = plotResults(y, anomalyScoresPCA, True,False) 
    print("n_components=" , i , " avg_precision=",avg_precision)

Scores are:

n_components= 10  avg_precision= 0.2845377775949285
n_components= 11  avg_precision= 0.29364944111841085
n_components= 12  avg_precision= 0.28594075836267696
n_components= 13  avg_precision= 0.31238695971860264
n_components= 14  avg_precision= 0.33482491519829544
n_components= 15  avg_precision= 0.3171596014339625
n_components= 16  avg_precision= 0.36246230687638303
n_components= 17  avg_precision= 0.3551693238316446
n_components= 18  avg_precision= 0.3576986952085249
n_components= 19  avg_precision= 0.3541947874943925
n_components= 20  avg_precision= 0.37019433800859325
n_components= 21  avg_precision= 0.37590377049563356
n_components= 22  avg_precision= 0.38371878690273875
n_components= 23  avg_precision= 0.3819741346612291
n_components= 24  avg_precision= 0.39826963028826606
n_components= 25  avg_precision= 0.4292470766080597
n_components= 26  avg_precision= 0.5661816082602096
n_components= 27  avg_precision= 0.6937067887351663
n_components= 28  avg_precision= 0.002783436997939959
n_components= 29  avg_precision= 0.00910804526226634

from handson-unsupervised-learning.

Train set for anomaly detection with PCA about handson-unsupervised-learning HOT 4 CLOSED

Comments (4)

Related Issues (17)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent