In this tutorial, we are going to create and deploy a simple machine learning model.
We are going to create the model using Python and the tensorflow library. We will finish up by serving the model in a flask application as an API(Application Programming Interface).

Let's get to it.

First create a directory/folder where you are going to put all the code from this tutorial.

The model we are going to create is a pretty simple one. Given an inputx, we want to get an output2x+1. Pretty simple! We want to create a model that will train on some sample data and come up with its own method to be able to find the right output.
Our sample data is pretty simple one.

sample_input=(1,2,3,4,5,6)sample_output=(3,5,7,9,11,13)

If you try using the formula on the input, you should get a corresponding output.

Now open the project folder in your favorite editor and let's get to coding.

Creating the model

Create a new file calledmodel.py and we are going ro write our model code in there.

First we are going to import tensorflow and keras.

importtensorflowastffromtensorflowimportkeras

Next is to create out model instance. Since it is simple one, a sequential model will be fine.

model=keras.Sequential([keras.layers.Dense(1,activation="relu",input_shape=[1])])

Next step is to compile our model and give it an optmizer as well as a loss function. The loss function will help to look back at how the model progresses from errors while rhe optmizer will help optmize the model.

model.compile(optimizer="sgd",loss="mean_squared_error")

Next comes training our model.

# train modelmodel.fit(sample_input,sample_output,epochs=500)

You should now end up with this

importtensorflowastffromtensorflowimportkerasmodel=keras.Sequential([keras.layers.Dense(1,activation="relu",input_shape=[1])])sample_input=(1,2,3,4,5,6)sample_output=(3,5,7,9,11,13)model.compile(optimizer="sgd",loss="mean_squared_error")# train modelmodel.fit(sample_input,sample_output,epochs=500)

Time to test our model. Add this line at the end of the file. You can pass any value in it to see if the model works. Make sure its in the form of a list.

print(model.predict([8]))# my output [[17.074446]]

After training the model, we try to make it predict on a sample value. I choosed to test on 8 as my sample and i got ~17. If we substitute in our formula, 2(8)+1 equal 17. This confirms that our model actually works.

Saving the model

After creating the model, it is good to save it so it can be deployed more easily.
Saving the model is as simple as using themodel.save() method.

Add this line at the end of your model.py file.

model.save(filepath="./")

The model.save() method takes afilepath keyword argument which is the path to which you want to save the model. I am saving my model in the root of the project.
You should see a file namedsaved_model.pb. This is our model and can now be shared or deployed.

Deploying the model

We are now going to create a simple flask API to serve our model.

Create a new file namedapp.py and add the follwing code in there.

fromtensorflow.kerasimportmodelsfromflaskimportFlask,requestapp=Flask(__name__)# load saved model from filesystemmodel=models.load_model(filepath="./")@app.route('/predict',methods=["POST"])defpredict():value=int(request.data)prediction=model.predict([value])# print(prediction)returnstr(prediction[0][0])if__name__=="__main__":app.run(debug=True)

First, we are importing models utility from keras and then Flask.
We can load a saved model using the models utility using the load_model method. It takes afilepath argument specifying the path to which the saved model is.

Next is we create a route which accpetPOST requests.
Inside the function, we get the data sent through the request. We will get bytes, so we cast it to an integer.
We then we used our loaded model to predict the output of the value sent. We then return our prediction. Notice we cast the response a string. This is because flask allows us to return only strings, tuples or dictionaries.

Testing our API

To test your API, we are going to write a simple python program to do so.
Open a file calledtest.py and add the following code in it.

importrequestsimportsysvalue=sys.argv[1]res=requests.post("http://localhost:5000/predict",value)print(res.content)

Testing it

python test.py 9

I getb'19.115118'. Yay it works!

You have reached the end of this tutorial.
I hope you enjoyed building and serving this simple model.

Feel free to play with the API with different values.

Final codehere

If you liked this, make sure to share andfollow me on twitter for more of such.

Other articles of mine that you might like

• The difference between String and string in Typescript
• Scooter: Write and publish short code snippets
• Create your own HTML element.
• Some JavaScript array methods
• My hashnode Christmas Hackathon submition.