AI Deploy - Tutoriel - Déployer une app interactive pour l'EDA et la prédiction en utilisant Streamlit (EN)

Objective

The objective of this tutorial is to deploy an application for Exploratory Data Analysis (EDA) and interactive prediction.

The use case is the IRIS dataset. It is a collection of data containing information on 50 observations of four different variables: petal length, petal width, sepal length and sepal width.

The first objective is to perform data mining on the source data from the Iris dataset.

Then, the goal is to make predictions using a trained model. We will therefore use a neural network and weights from this notebook to classify iris flowers between different species (Setosa, Versicolor and Virginica). The four features are sent as input to the neural network.

The PyTorch model will be loaded and run, and the prediction probabilities for each class will be extracted. Depending on the input data, it will be possible to visualize the corresponding iris on a PCA plot to compare the input to other data points in our Iris dataset.

In order to do this, you will use Streamlit, a Python framework that turns scripts into a shareable web application. You will also learn how to build and use a custom Docker image for a Streamlit application.

Requirements

• Access to the OVHcloud Control Panel;
• An AI Deploy project created inside a Public Cloud project in your OVHcloud account;
• A user for AI Deploy;
• Docker installed on your local computer;
• Some knowledge about building image and Dockerfile;
• You also should have followed the Step 6 of the Classification of iris varieties within the same species notebook from the GitHub repository. You will be able to save your PyTorch model model_iris_classification.pth.

Instructions

You are going to follow different steps to build your Streamlit application.

• More information about Streamlit capabilities can be found here.
• Direct link to the full Python file can be found here here.

Here we will mainly discuss how to write the utils.py and app.py codes, the requirements.txt file and the Dockerfile. If you want to see the whole code, please refer to the GitHub repository.

Create the Streamlit application

Two Python files are created for the purpose of loading the PyTorch model and building the Streamlit app.

Define the utils.py file

This Python file is dedicated to load the PyTorch model. You can find the full code here.

First, we have to define the Model class with neural network architecture.

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        # fully connected layer : 4 input features for 4 parameters in X
        self.layer1 = nn.Linear(in_features=4, out_features=16)
        # fully connected layer
        self.layer2 = nn.Linear(in_features=16, out_features=12)
        # output layer : 3 output features for 3 species
        self.output = nn.Linear(in_features=12, out_features=3)

    def forward(self, x):
        # activation fonction : reLU
        x = F.relu(self.layer1(x))
        x = F.relu(self.layer2(x))
        x = self.output(x)
        return x

Then, we create the function to load model checkpoint.

def load_checkpoint(path):

    model = Model()
    print("Model display: ", model)
    model.load_state_dict(torch.load(path))
    model.eval()

    return model

The last function allows us to load the model and get predictions.

def load_model(X_tensor):
    model = load_checkpoint(path)
    predict_out = model(X_tensor)
    _, predict_y = torch.max(predict_out, 1)
    return predict_out.squeeze().detach().numpy(), predict_y.item()

# pytorch model
path = "model_iris_classification.pth"

Write the app.py file

Load the IRIS dataset for EDA.

@st.cache
def load_data():
    # load the iris dataset with sklearn
    dataset_iris = load_iris()
    # define inputs and output
    df_inputs = pd.DataFrame(dataset_iris.data, columns=dataset_iris.feature_names)
    df_output = pd.DataFrame(dataset_iris.target, columns=['variety'])
    return df_inputs, df_output

Display EDA figure based on source dataset.

@st.cache(allow_output_mutation=True)
def data_visualization(df_inputs, df_output):
    df = pd.concat([df_inputs, df_output['variety']], axis=1)
    eda = sns.pairplot(data=df, hue="variety", palette=['#0D0888', '#CB4779', '#F0F922'])
    return eda

Create a sidebar with sliders.

def create_slider(df_inputs):
    # slidebars with min, max and mean (by default) values
    sepal_length = st.sidebar.slider(
        label='Sepal Length',
        min_value=float(df_inputs['sepal length (cm)'].min()),
        max_value=float(df_inputs['sepal length (cm)'].max()),
        value=float(round(df_inputs['sepal length (cm)'].mean(), 1)),
        step=0.1)
    sepal_width = st.sidebar.slider(
        label='Sepal Width',
        min_value=float(df_inputs['sepal width (cm)'].min()),
        max_value=float(df_inputs['sepal width (cm)'].max()),
        value=float(round(df_inputs['sepal width (cm)'].mean(), 1)),
        step=0.1)
    petal_length = st.sidebar.slider(
        label='Petal Length',
        min_value=float(df_inputs['petal length (cm)'].min()),
        max_value=float(df_inputs['petal length (cm)'].max()),
        value=float(round(df_inputs['petal length (cm)'].mean(), 1)),
        step=0.1)
    petal_width = st.sidebar.slider(
        label='Petal Width',
        min_value=float(df_inputs['petal width (cm)'].min()),
        max_value=float(df_inputs['petal width (cm)'].max()),
        value=float(round(df_inputs['petal width (cm)'].mean(), 1)),
        step=0.1)
    return sepal_length, sepal_width, petal_length, petal_width

Run a PCA.

@st.cache
def run_pca():
    pca = PCA(2)
    X = df_inputs.iloc[:, :4]
    X_pca = pca.fit_transform(X)
    df_pca = pd.DataFrame(pca.transform(X))
    df_pca.columns = ['PC1', 'PC2']
    df_pca = pd.concat([df_pca, df_output['variety']], axis=1)
    return pca, df_pca

Create a function that filters out negative values from the dataframe. Only positive values in the dataframe are kept. If a value is negative, it is set to zero.

def extract_positive_value(prediction):
    # f(prediction) = max(0, prediction)
    prediction_positive = []
    for p in prediction:
        if p < 0:
            p = 0
        prediction_positive.append(p)
    return pd.DataFrame({'Species': ['Setosa', 'Versicolor', 'Virginica'], 'Confidence': prediction_positive})

Define a Python function to display image according to Iris species.

def display_img(species):
    # define the list of images
    list_img = ['setosa.png', 'versicolor.png', 'virginica.png']
    return Image.open(list_img[species])

Write the requirements.txt file for the application

The requirements.txt file will allow us to write all the modules needed to make our application work. This file will be useful when writing the Dockerfile.

pandas==1.4.4
numpy==1.23.2
torch==1.12.1
streamlit==1.12.2
scikit-learn==1.1.2
plotly==5.10.0
Pillow==9.2.0
seaborn==0.12.0

Write the Dockerfile for the application

Your Dockerfile should start with the FROM instruction indicating the parent image to use. In our case we choose to start from the python:3.8 OVHcloud image:

python:3.8

Create the home directory and add your files to it:

WORKDIR /workspace
ADD . /workspace

Install the requirements.txt file which contains your needed Python modules using a pip install ... command:

RUN pip install -r requirements.txt

Define your default launching command to start the application:

CMD [ "streamlit", "run", "/workspace/app.py", "--server.address=0.0.0.0" ]

Give correct access rights to ovhcloud user (42420:42420):

RUN chown -R 42420:42420 /workspace
ENV HOME=/workspace

Build the Docker image from the Dockerfile

From the directory containing your Dockerfile, run one of the following commands to build your application image:

# Build the image using your machine's default architecture
docker build . -t streamlit-eda-iris:latest

# Build image targeting the linux/amd64 architecture
docker buildx build --platform linux/amd64 -t streamlit-eda-iris:latest .

• The first command builds the image using your system’s default architecture. This may work if your machine already uses the linux/amd64 architecture, which is required to run containers with our AI products. However, on systems with a different architecture (e.g. ARM64 on Apple Silicon), the resulting image will not be compatible and cannot be deployed.

• The second command explicitly targets the linux/AMD64 architecture to ensure compatibility with our AI services. This requires buildx, which is not installed by default. If you haven’t used buildx before, you can install it by running: docker buildx install

Test it locally (optional)

Launch the following Docker command to launch your application locally on your computer:

docker run --rm -it -p 8501:8501 --user=42420:42420 streamlit-eda-iris:latest

Once started, your application should be available on http://localhost:8501.

Push the image into the shared registry

Find the address of your shared registry by launching this command:

ovhai registry list

Log in on the shared registry with your usual AI Platform user credentials:

docker login -u <user> -p <password> <shared-registry-address>

Push the compiled image into the shared registry:

docker tag streamlit-eda-iris:latest <shared-registry-address>/streamlit-eda-iris:latest
docker push <shared-registry-address>/streamlit-eda-iris:latest

Launch the AI Deploy app

The following command starts a new app running your Streamlit application:

ovhai app run \
      --default-http-port 8501 \
      --cpu 12 \
      <shared-registry-address>/streamlit-eda-iris:latest

Go further

• You can imagine deploying an AI model with an other tool: Gradio. Refer to this tutorial.
• Another way to create an AI Deploy app is to use Flask! Here it is.

If you need training or technical assistance to implement our solutions, contact your sales representative or click on this link to get a quote and ask our Professional Services experts for a custom analysis of your project.

Feedback

Please send us your questions, feedback and suggestions to improve the service:

• On the OVHcloud Discord server