MLOps for MLE: Model Development

Posted on In

Base Model Development

학습 및 평가 데이터 선정

1
2
3
4
5
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split

X, y = load_breast_cancer(return_X_y=True, as_frame=True)
X_train, X_valid, y_train, y_valid = train_test_split(X, y, train_size=0.8, random_state=2023)

모델 개발 및 학습

1
2
3
4
5
from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
scaled_X_train = scaler.fit_transform(X_train)
scaled_X_valid = scaler.transform(X_valid)
  • sklearn.preprocessing.StandardScaler(): Standardize features by removing the mean and scaling to unit variance
    • $z = (x - u) / s$
      • $u$: Mean of the training samples
      • $s$: Standard deviation of the training samples
    • StandardScaler.fit_transform(X): Fit to data, then transform it
    • StandardScaler.transform(X): Perform standardization by centering and scaling
1
2
3
4
5
6
7
8
9
10
11
12
13
14
from sklearn.svm import SVC
from sklearn.metrics import accuracy_score

classifier = SVC()
classifier.fit(scaled_X_train, y_train)

train_pred = classifier.predict(scaled_X_train)
valid_pred = classifier.predict(scaled_X_valid)

train_acc = accuracy_score(y_true=y_train, y_pred=train_pred)
valid_acc = accuracy_score(y_true=y_valid, y_pred=valid_pred)

print("Train Accuracy [%]:", train_acc * 100) # Train Accuracy [%]: 98.24175824175823
print("Valid Accuracy [%]:", valid_acc * 100) # Valid Accuracy [%]: 97.36842105263158
  • sklearn.svm.SVC(): C-Support Vector Classification
    • SVC.fit(X, y): Fit the SVM model according to the given training data
    • SVC.predict(X): Number of support vectors for each class
  • sklearn.metrics.accuracy_score(y_true, y_pred): Accuracy classification score

모델 파이프라인

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
from sklearn.metrics import accuracy_score
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.svm import SVC

model_pipeline = Pipeline([("scaler", StandardScaler()), ("svc", SVC())])
model_pipeline.fit(X_train, y_train)

train_pred = model_pipeline.predict(X_train)
valid_pred = model_pipeline.predict(X_valid)

train_acc = accuracy_score(y_true=y_train, y_pred=train_pred)
valid_acc = accuracy_score(y_true=y_valid, y_pred=valid_pred)

print("Train Accuracy :", train_acc) # Train Accuracy : 0.9875
print("Valid Accuracy :", valid_acc) # Valid Accuracy : 0.95
  • sklearn.pipeline.Pipeline(steps): Pipeline of transforms with a final estimator
    • Pipeline.fit(X, y): Fit the model
    • Pipeline.predict(X): Transform the data, and apply predict with the final estimator

학습된 모델 저장, 저장된 모델 불러오기

1
2
3
import joblib

joblib.dump(model_pipeline, "Etc./db_pipeline.joblib")
  • joblib.dump(value, filename): Persist an arbitrary Python object into one file
1
pipeline_load = joblib.load("Etc./db_pipeline.joblib")
  • joblib.load(filename): Reconstruct a Python object from a file persisted with joblib.dump

Load Data from Database

1
2
3
4
5
6
7
8
9
import pandas as pd
import psycopg2

# PGPASSWORD=qwer123! psql -h localhost -p 1234 -U zerohertz -d Breast_Cancer
db_connect = psycopg2.connect(host="localhost", database="Breast_Cancer", user="zerohertz", password="qwer123!", port="1234")
df = pd.read_sql("SELECT * FROM breast_cancer_data ORDER BY id DESC LIMIT 100", db_connect)
X, y = df.drop(["id", "timestamp", "target"], axis="columns"), df["target"]

df.to_csv("Etc./DB.csv", index=False)
  • psycopg2.connect(dbname, user, password)
    • dbname: The database name (database is a deprecated alias)
    • user: User name used to authenticate
    • password: Password used to authenticate
    • host: Database host address (defaults to UNIX socket if not provided)
    • port: Connection port number (defaults to 5432 if not provided)
  • pandas.read_sql(sql, con): Read SQL query or database table into a pandas.DataFrame
  • pandas.DataFrame.drop(label, axis): Drop specified labels from rows or columns
  • pandas.DataFrame.to_csv(path_or_buf, index=True): Write object to a comma-separated values (csv) file

PostgreSQL DB 기반 모델 훈련

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
import joblib
import pandas as pd
import psycopg2
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.svm import SVC

# PGPASSWORD=qwer123! psql -h localhost -p 1234 -U zerohertz -d Breast_Cancer
db_connect = psycopg2.connect(host="localhost", database="Breast_Cancer", user="zerohertz", password="qwer123!", port="1234")
df = pd.read_sql("SELECT * FROM breast_cancer_data ORDER BY id DESC LIMIT 100", db_connect)
X, y = df.drop(["id", "timestamp", "target"], axis="columns"), df["target"]
X_train, X_valid, y_train, y_valid = train_test_split(X, y, train_size=0.8, random_state=2023)

model_pipeline = Pipeline([("scaler", StandardScaler()), ("svc", SVC())])
model_pipeline.fit(X_train, y_train)

train_pred = model_pipeline.predict(X_train)
valid_pred = model_pipeline.predict(X_valid)

train_acc = accuracy_score(y_true=y_train, y_pred=train_pred)
valid_acc = accuracy_score(y_true=y_valid, y_pred=valid_pred)

print("Train Accuracy :", train_acc) # Load Model Train Accuracy : 0.9875
print("Valid Accuracy :", valid_acc) # Load Model Valid Accuracy : 0.95

joblib.dump(model_pipeline, "Etc./db_pipeline.joblib")

df.to_csv("Etc./DB.csv", index=False)

모델 검증

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
import joblib
import pandas as pd
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split

df = pd.read_csv("Etc./DB.csv")
X, y = df.drop(["id", "timestamp", "target"], axis="columns"), df["target"]
X_train, X_valid, y_train, y_valid = train_test_split(X, y, train_size=0.8, random_state=2023)

pipeline_load = joblib.load("Etc./db_pipeline.joblib")

load_train_pred = pipeline_load.predict(X_train)
load_valid_pred = pipeline_load.predict(X_valid)

load_train_acc = accuracy_score(y_true=y_train, y_pred=load_train_pred)
load_valid_acc = accuracy_score(y_true=y_valid, y_pred=load_valid_pred)

print("Load Model Train Accuracy :", load_train_acc) # Load Model Train Accuracy : 0.9875
print("Load Model Valid Accuracy :", load_valid_acc) # Load Model Valid Accuracy : 0.95

Reference

  • MLOps for MLE: 02. Model Development