Skip to content

Section 2: MLflow Tracking

Simple ML model -

We have implemented a simple ML model to showcase the experiment tracking concept used in MLflow-

Source code

simple_ML_model.py 
 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
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
import os
import argparse
import pandas as pd
import numpy as np

from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.model_selection import train_test_split
from sklearn.linear_model import ElasticNet
from urllib.parse import urlparse
import mlflow
import mlflow.sklearn


def get_data():
    URL = "http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv"

    try:
        df = pd.read_csv(URL, sep=";")
        return df
    except Exception as e:
        raise e

def evaluate(actual, pred):
    rmse = np.sqrt(mean_squared_error(actual, pred))
    mae = mean_absolute_error(actual, pred)
    r2 = r2_score(actual, pred)
    return rmse, mae, r2

def main(alpha, l1_ratio):

    df = get_data()

    train, test = train_test_split(df)

    train_x = train.drop(["quality"], axis=1)
    test_x = test.drop(["quality"], axis=1)

    train_y = train[["quality"]]
    test_y = test[["quality"]]

    # mlflow 
    with mlflow.start_run():
        lr = ElasticNet(alpha=alpha, l1_ratio=l1_ratio, random_state=42)
        lr.fit(train_x, train_y)

        pred = lr.predict(test_x)

        rmse, mae, r2 = evaluate(test_y, pred)

        print(f"Elastic net params: alpha: {alpha}, l1_ratio: {l1_ratio}")
        print(f"Elastic net metric: rmse:{rmse}, mae: {mae}, r2:{r2}")

        mlflow.log_param("alpha", alpha)
        mlflow.log_param("l1_ratio", l1_ratio)

        mlflow.log_metric("rmse", rmse)
        mlflow.log_metric("mae", mae)
        mlflow.log_metric("r2", r2)


if __name__=="__main__":
    args = argparse.ArgumentParser()
    args.add_argument("--alpha", "-a", type=float, default=0.5)
    args.add_argument("--l1_ratio", "-l1", type=float, default=0.5)
    parsed_args = args.parse_args()
    try:
        main(alpha=parsed_args.alpha, l1_ratio=parsed_args.l1_ratio)
    except Exception as e:
        raise e

Concept of Runs

MLflow Tracking is based on runs. Runs are executions of some piece of data science code. A Run can record the following :

  • Code Version
  • Start & End Time
  • Source
  • Parameters
  • Metrics
  • Artifacts

Logging our simple ML model using

In this lecture it has been shown that how we can log our model for every execution or experiment-

Source code

simple_ML_model_2.py 
 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
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
import os
import argparse
import pandas as pd
import numpy as np

from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.model_selection import train_test_split
from sklearn.linear_model import ElasticNet
from urllib.parse import urlparse
import mlflow
import mlflow.sklearn


def get_data():
    URL = "http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv"

    try:
        df = pd.read_csv(URL, sep=";")
        return df
    except Exception as e:
        raise e

def evaluate(actual, pred):
    rmse = np.sqrt(mean_squared_error(actual, pred))
    mae = mean_absolute_error(actual, pred)
    r2 = r2_score(actual, pred)
    return rmse, mae, r2

def main(alpha, l1_ratio):

    df = get_data()

    train, test = train_test_split(df)

    train_x = train.drop(["quality"], axis=1)
    test_x = test.drop(["quality"], axis=1)

    train_y = train[["quality"]]
    test_y = test[["quality"]]

    # mlflow 
    with mlflow.start_run():
        lr = ElasticNet(alpha=alpha, l1_ratio=l1_ratio, random_state=42)
        lr.fit(train_x, train_y)

        pred = lr.predict(test_x)

        rmse, mae, r2 = evaluate(test_y, pred)

        print(f"Elastic net params: alpha: {alpha}, l1_ratio: {l1_ratio}")
        print(f"Elastic net metric: rmse:{rmse}, mae: {mae}, r2:{r2}")

        mlflow.log_param("alpha", alpha)
        mlflow.log_param("l1_ratio", l1_ratio)

        mlflow.log_metric("rmse", rmse)
        mlflow.log_metric("mae", mae)
        mlflow.log_metric("r2", r2)

        # mlflow model logging 
        mlflow.sklearn.log_model(lr, "model")


if __name__=="__main__":
    args = argparse.ArgumentParser()
    args.add_argument("--alpha", "-a", type=float, default=0.5)
    args.add_argument("--l1_ratio", "-l1", type=float, default=0.5)
    parsed_args = args.parse_args()
    try:
        main(alpha=parsed_args.alpha, l1_ratio=parsed_args.l1_ratio)
    except Exception as e:
        raise e

Exploring UI of MLflow

runs.py 
import numpy as np
import os

alpha_s=np.linspace(0.1, 1.0, 5)
l1_ratios=np.linspace(0.1, 1.0, 5)

for alpha in alpha_s:
for l1 in l1_ratios:
    os.system(f"python simple_ML_model_2.py -a {alpha} -l1 {l1}")

Info

Refer video lecture for this in oneNeuron platform for UI exploration

Packaging a project MLflow way

  • Create a conda.yaml file as shown below:

    conda.yaml 
    name: mlflow_tutorial
channels:
    - defaults
dependencies:
    - python=3.7.11=h6244533_0
    - pip=21.2.4=py37haa95532_0
    - pip:
        - mlflow==1.23.1
        - numpy==1.21.5
        - pandas==1.3.5
        - scikit-learn==1.0.2

  • or run the following command to create conda.yaml file 

    conda env export > conda.yaml

    Note

    make sure you are in the same environment while running the command whose conda.yaml file you wish to create

  • after above step create the an MLproject file in the root of the project as shown below -

    MLproject 
    name: mlflow_tutorial

conda_env: conda.yaml

entry_points:
    main:
        parameters:
            alpha: {type: float, default: 0.5}
            l1_ratio: {type: float, default: 0.5}
        command: "python simple_ML_model_2.py -a {alpha} -l1 {l1_ratio}"

  • Now run the following command to execte the project

    • without using a fresh conda environment by using the existing environment- 
      mlflow run . --no-conda
    • with a fresh conda environment - 
      mlflow run .
    • if you wish to pass command line argument then use the below command- 
      mlflow run . -P alpha=0.7 -P l1_ratio=0.4

  • source code for the above demo - source code

MLflow tracking server

  • make the changes in the code base to be ready for tracking server -

    code 
     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
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
    import os
import argparse
import pandas as pd
import numpy as np

from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.model_selection import train_test_split
from sklearn.linear_model import ElasticNet
from urllib.parse import urlparse
import mlflow
import mlflow.sklearn


def get_data():
    URL = "http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv"

    try:
        df = pd.read_csv(URL, sep=";")
        return df
    except Exception as e:
        raise e

def evaluate(actual, pred):
    rmse = np.sqrt(mean_squared_error(actual, pred))
    mae = mean_absolute_error(actual, pred)
    r2 = r2_score(actual, pred)
    return rmse, mae, r2

def main(alpha, l1_ratio, tracking_uri, port):

    df = get_data()

    train, test = train_test_split(df)

    train_x = train.drop(["quality"], axis=1)
    test_x = test.drop(["quality"], axis=1)

    train_y = train[["quality"]]
    test_y = test[["quality"]]

    # mlflow tracking URI
    URI = f"http://{tracking_uri}:{port}"
    mlflow.set_tracking_uri(URI)

    # mlflow 
    with mlflow.start_run():
        lr = ElasticNet(alpha=alpha, l1_ratio=l1_ratio, random_state=42)
        lr.fit(train_x, train_y)

        pred = lr.predict(test_x)

        rmse, mae, r2 = evaluate(test_y, pred)

        print(f"Elastic net params: alpha: {alpha}, l1_ratio: {l1_ratio}")
        print(f"Elastic net metric: rmse:{rmse}, mae: {mae}, r2:{r2}")

        mlflow.log_param("alpha", alpha)
        mlflow.log_param("l1_ratio", l1_ratio)

        mlflow.log_metric("rmse", rmse)
        mlflow.log_metric("mae", mae)
        mlflow.log_metric("r2", r2)

        # register model in the sql server
        mlflow.sklearn.log_model(lr, "model", registered_model_name="ENmodel")


if __name__=="__main__":
    args = argparse.ArgumentParser()
    args.add_argument("--alpha", "-a", type=float, default=0.5)
    args.add_argument("--l1_ratio", "-l1", type=float, default=0.5)
    args.add_argument("--tracking_uri", "-t", type=str, default="localhost")
    args.add_argument("--port", "-p", type=int, default=5000)
    parsed_args = args.parse_args()
    try:
        main(alpha=parsed_args.alpha, l1_ratio=parsed_args.l1_ratio,
        tracking_uri=parsed_args.tracking_uri, port=parsed_args.port)

    except Exception as e:
        raise e

  • run mlflow sqlite server to store parameters and metrics in an sqlite local database and create artifact directory to store files 

    mlflow server \
--backend-store-uri sqlite:///mlflow.db \
--default-artifact-root ./artifacts \
--host 0.0.0.0 -p 1234
    Info

    To create command using conda env file [an alternative to pip install -r requirements.txt]

    conda env create --prefix ./env -f conda.yaml
Back to top