In this tutorial, we will walk through the steps of deploying a machine learning model as an API using FastAPI in Python. FastAPI is a modern, fast (high-performance), web framework for building APIs with Python 3.6+ based on standard Python type hints.
Before we begin, make sure you have the following prerequisites:
- Basic knowledge of Python programming.
- Understanding of machine learning concepts and how to build and train a model.
- Familiarity with creating and working with APIs.
Let’s get started:
Step 1: Build and train your machine learning model
First, you need to build and train your machine learning model using any library/framework of your choice (e.g. Scikit-learn, TensorFlow, PyTorch). For the purpose of this tutorial, we will use a simple example of building a linear regression model using Scikit-learn.
Here is an example code snippet for building and training a linear regression model:
from sklearn.linear_model import LinearRegression
import pandas as pd
# Load data
data = pd.read_csv('data.csv')
X = data[['feature1', 'feature2']]
y = data['target']
# Initialize and train the model
model = LinearRegression()
model.fit(X, y)Step 2: Create an API using FastAPI
Next, we will create an API using FastAPI to deploy our machine learning model.
First, install FastAPI and Uvicorn (a lightning-fast ASGI server) using pip:
pip install fastapi uvicornNext, create a new Python file (e.g. app.py) and import FastAPI:
from fastapi import FastAPI
import numpy as np
app = FastAPI()Step 3: Define API endpoints
Next, define the API endpoints that will serve as the interface for interacting with the model.
For example, we can create a /predict endpoint that takes input features as JSON data and returns the prediction from the model:
@app.get("/predict")
def predict(feature1: float, feature2: float):
features = np.array([[feature1, feature2]])
prediction = model.predict(features)
return {"prediction": prediction[0]}Step 4: Run the API server
Finally, run the API server using Uvicorn:
uvicorn app:app --reloadNow, you can access the API endpoint at http://127.0.0.1:8000/predict and send a POST request with input data to get predictions from the machine learning model.
That’s it! You have successfully deployed a machine learning model as an API using FastAPI in Python. You can customize and extend the API as needed to support more endpoints and functionality.
I hope this tutorial was helpful in guiding you through deploying your machine learning model as an API. Happy coding!
Hi everyone,
I encountered the same issue where the model consistently predicted non-diabetic cases the same way. I resolved it by making the following adjustment, which worked for me:
### Data Standardization:
In our training code, we used `StandardScaler` to standardize the data. When deploying the model with FastAPI, it's crucial to ensure that the input data is standardized in the same way before making predictions.
To fix this, I added the `StandardScaler` transformation to our FastAPI code, using the same scaler that was fitted during training. Here’s what the `api.py` file looks like:
“`python
import requests
url = 'http://127.0.0.1:8000/diabetes_prediction'
input_data_for_model = {
'pregnancies': 4,
'Glucose': 110,
'BloodPressure': 92,
'SkinThickness': 0,
'Insulin': 0,
'BMI': 37.6,
'DiabetesPedigreeFunction': 0.191,
'Age': 30
}
response = requests.post(url, json=input_data_for_model)
print(response.json())
“`
And here’s the `ml.py` code:
“`python
import json
from fastapi import FastAPI
from pydantic import BaseModel
import pickle
import numpy as np
app = FastAPI()
class ModelInput(BaseModel):
pregnancies: int
Glucose: int
BloodPressure: int
SkinThickness: int
Insulin: int
BMI: float
DiabetesPedigreeFunction: float
Age: int
# Load the model and the scaler
diabetes_model = pickle.load(open('diabetes_model.sav', 'rb'))
scaler = pickle.load(open('scaler.sav', 'rb'))
@app.post('/diabetes_prediction')
def diabetes_pred(input_parameters: ModelInput):
input_data = np.array([
input_parameters.pregnancies,
input_parameters.Glucose,
input_parameters.BloodPressure,
input_parameters.SkinThickness,
input_parameters.Insulin,
input_parameters.BMI,
input_parameters.DiabetesPedigreeFunction,
input_parameters.Age
])
input_data_reshaped = input_data.reshape(1, -1)
std_data = scaler.transform(input_data_reshaped)
prediction = diabetes_model.predict(std_data)
if prediction[0] == 0:
return {'prediction': 'The person is not diabetic'}
else:
return {'prediction': 'The person is diabetic'}
“`
Additionally, I made a small change when saving the model:
“`python
import pickle
# Save the model
filename = 'diabetes_model.sav'
pickle.dump(classifier, open(filename, 'wb'))
# Save the scaler
scaler_filename = 'scaler.sav'
pickle.dump(scaler, open(scaler_filename, 'wb'))
“`
I hope this helps you all!
Anna can you do project on Genetic Disorder Prediction plz…
True GEM, MAN //\
sir will this code will generate API even if our model is not able to run in python shell it is only able to run on google collab because of large training data and hardware restrictions
Nice implementation
I'm getting 'Internal server error' , pls help
great job, thanks
You are great teacher, thanks a million) Very rare decent corse!
What if the model is cnn how i can do it ?
Hello Sir
I'm facing this problem
raise PydanticUserError(
pydantic.errors.PydanticUserError: A non-annotated attribute was detected: `diabetes_model = SVC(kernel='linear')`. All model fields require a type annotation; if `diabetes_model` is not meant to be a field, you may be able to resolve this error by annotating it as a `ClassVar` or updating `model_config['ignored_types']`.
Thank you.This is a greate explanation.!!One of the best video of this category….
Thank you for the informative content bro 👍, it was very helpfull.
Siddhardhan, you have nicely explained the topic. Thank you.
Amazing video. Well done
best ml tutorial i have ever seen .thank you brother.
sir I want a deploy model for spam mail can you help me please
I tried by watching this video but it is throwing the error please help me
I've to present it next week
Why we need to pass as json why not dictionary only
Sir on testing it I am getting "self.probA_, self.probB_, svm_type=svm_type, kernel=kernel,
AttributeError: 'SVC' object has no attribute 'probA_' " error. How I can resolve that?
Amazing video.Thank you
I have a problem
I couldn't install pickle5