Help Center/ ModelArts/ Model Inference/ Inference Specifications/ Examples of Custom Scripts/ Scikit-learn
Updated on 2024-06-12 GMT+08:00
View PDF

Scikit-learn

Training and Saving a Model

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
 
import json
import pandas as pd
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.externals import joblib
iris = pd.read_csv('/home/ma-user/work/iris.csv')
X = iris.drop(['variety'],axis=1)
y = iris[['variety']]
# Create a LogisticRegression instance and train model
logisticRegression = LogisticRegression(C=1000.0, random_state=0)
logisticRegression.fit(X,y)
# Save model to local path
joblib.dump(logisticRegression, '/tmp/sklearn.m')

Before training, download the iris.csv dataset, decompress it, and upload it to the /home/ma-user/work/ directory of the notebook instance. Download the iris.csv dataset from https://gist.github.com/netj/8836201. For details about how to upload a file to a notebook instance, see Upload Scenarios and Entries.

After the model is saved, it must be uploaded to the OBS directory before being published. The config.json and customize_service.py files must be contained during publishing. For details about the definition method, see Introduction to Model Package Specifications.

Inference Code

In the model inference code file customize_service.py, add a child model class. This child model class inherits properties from its parent model class. For details about the import statements of different types of parent model classes, see Table 1.

 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
31
32
 
# coding:utf-8
import collections
import json
from sklearn.externals import joblib
from model_service.python_model_service import XgSklServingBaseService

class UserService(XgSklServingBaseService):

    # request data preprocess
    def _preprocess(self, data):
        list_data = []
        json_data = json.loads(data, object_pairs_hook=collections.OrderedDict)
        for element in json_data["data"]["req_data"]:
            array = []
            for each in element:
                array.append(element[each])
                list_data.append(array)
        return list_data

    # predict
    def _inference(self, data):
        sk_model = joblib.load(self.model_path)
        pre_result = sk_model.predict(data)
        pre_result = pre_result.tolist()
        return pre_result

    # predict result process
    def _postprocess(self,data):
        resp_data = []
        for element in data:
            resp_data.append({"predictresult": element})
        return resp_data