Building Deep Neural Networks the Easy Way | Perceptilabs

Building Deep Neural Networks the Easy Way | Perceptilabs

A tutorial to build a DNN on Covid 19 patients Xray scans using Perceptilabs


7 min read

PerceptiLabs' visual simulation model offers a graphical user interface for creating, learning, and evaluating designs as well as allowing for further programming modifications. You can get quick repetitions and improved solutions that are easier to describe.

The framework of Perceptible allows users to create modified model configurations without requiring scientific knowledge as well as end-to-end simulation techniques that enable users to perceive and analyze the model in an entirely clear way improving awareness and allowing for error detection.

What is Perceptilabs?

This is actually a user interface for TensorFlow with an advanced machine learning platform with a graphical modeling process that combines the freedom of programming with the convenience of a drag-and-drop interface that would be a visual design on top of TensorFlow. This can make modeling creation simpler, quicker, and more available to a broader range of people.

It also includes pre-built algorithms for a variety of disciplines that can be transferred into the workplace by individuals, enabling them to modify and learn these on their own datasets. Tax fraud identification, object classification for pattern recognition, and other applications are among the system's use applications.

PerceptiLabs and machine learning

PerceptiLabs was founded with the goal of making machine learning modeling easier for businesses of all sizes. Machine learning could have a vital aspect in our development, and PerceptiLabs is now on a journey to enable businesses of all sizes to get started in this specific industry.

It analyses the ever-increasing amounts of data accessible today, assists businesses in identifying trends in the information, and provides estimates depending on those trends. Every business has a range of applications, such as employing object identification to predict which grocery stores are getting low on stock or utilizing picture identification to recognize a person in a congested field.

Users can easily create machine learning algorithms for any type of business with PerceptiLab's visual modeling solution. It enables users to click and drag items and join elements, then configures variables before the software writes their programming instantly. Users may quickly train and fine-tune their machine learning model, as well as observe its performance.

Modeling workflow of Perceptilabs


Photo by Tobias Carlsson on Unsplash

Pre-made elements encapsulate TensorFlow data and simplify it into visible components, while still enabling customized code updates. This graphical interface enables you to move these elements into a structure that depicts the design of your system. This user interface makes it simple to implement additional features such as one-hot encoding and thick layering.

As you alter the design in PerceptiLabs, every element also offers graphical information on how it has converted the dataset. This immediate overview reduces the requirement to execute the entire simulation before viewing results, allowing you to change more quickly.

Whenever you compare PerceptiLabs to any other platform, you'll see how much easier it is to visualize pictures and categorize information. You could also observe how every element alters the information, as well as how the alterations contributed to the final categorization.

During modeling, PerceptiLabs retrieves and utilizes the initial piece of an accessible dataset, and it re-runs the system as you implement adjustments, and you'll see how your modifications affect your outcome right away. This useful tool allows you to examine results without having to execute the algorithm on the whole sample.

Building your first Deep Learning model on Perceptilabs

If you are interested in a more comprehensive video tutorial, check out my youtube video below

Step 1: Install and run Perceptilabs on local Open terminal, use pip to install and run the tool locally (Make sure to have python version < 3.9

pip install perceptilabs

After the setup the tool is up and running on localhost:8080 Step 2: Understanding the dataset I am using the default sample dataset of X-Ray scans of patients provided in Perceptilabs. The dataset has Xray scans with 3 labels - Normal,Viral Pneumonia, Covid image.png To import this dataset in Perceptilabs you would need it in the right format which is in data.csv file contains the path of the files with corresponding labels

Step 3: Go to the model hub (first tab on the left), click on create model and import the dataset data.csv

image.png Select the URL as input feature and labels as Target Keep the data partition as default [70% Train,20% Validation,10% Test]

Select the training settings

image.png Provide the details such as Model Name, Epochs, Batch Size, Loss Function, Learning Rate and click on Customize and go to the Modelling window

Within the modelling window you will see all the layers of the neural network being laid out as per the inference of the modeling tool, it should look something like the screen shot below

image.png It contains by default 1 convolution layer connected to input images and two dense layers with softmax to convert it to a final label output

Step 4: Play around with the tool and multiple layers You can add in more deep learning components as a part of modeling tool, or you can easily code a custom Keras function for another layer


Building any deep learning model requires a lot of iterations, therefore the visual approach comes in handy where we can plug play and see the iterated results

Step 5: Start Training and see live stats (Statistics View) Click on Run with current settings on the top bar to start training the model, pass in the model settings as discussed previously, for a classification use case a Cross-Entropy Loss function would make more sense.


After you start modelling, you will be redirected to the statistics view, to see the live statistics of the model while it is getting trained, you should be able to see weights output, loss, accuracy of each row of the dataset being trained. And all of this analysis can be done layer by layer image.png

You should also be able to see the accuracy increase as the epoch pass on a global level


Step 6: Run Validation on Tests Dataset Go to the test view and run the test to get model metrics and confusion matrix of labels


And after it is complete you should be able to see the quality of the model you have built using these test metrics such as Model Accuracy, Precision, Recall


Why something like Perceptilabs makes sense?

Data analysts may use this technology to perform more effectively with machine learning techniques and get a good understanding of them.

Helps you get the Real-time information

Real-time metrics and detailed summaries of every modeling element's data are available. You can simply follow and analyze the behavior of the variables, troubleshoot in real-time, and identify where your system may be improved.

Helps you share them on GitHub

PerceptiLabs allows you to maintain many simulations, evaluate them, and communicate the findings with your group quickly and efficiently. Export your data as a TensorFlow framework.

Helps you overcome Compatibility problems

When a corporation's researchers create models and put them into operation, they must all be using the same model. Otherwise, problems would arise. According to some experts, this problem may be avoided if everyone in a firm utilizes PerceptiLabs' technology. Helps you export your model

Perceptilabs allows you to examine and explain how your program runs and executes, as well as why particular outcomes are being produced. You may also export your data as a training TensorFlow version after you're okay with it.

Advantages of using Perceptilabs

This tool offers a wide range of benefits. Some of them are;

  • Quick modeling - Includes a simple drag-and-drop user interface that helps make system design simple to create and analyze.
  • Visibility - It can be used to start understanding how your strategy performs so that it can be explained.
  • Versatility - Built as a graphical API on top of TensorFlow, this allows programmers to use TensorFlow's low-level Interface while also allowing them to use other Python libraries.


The procedure of developing algorithms must be simplified if businesses are to embrace machine learning. PerceptiLabs offers graphical machine learning modeling solutions to assist businesses in implementing computer learning. It not only allows you to develop computer learning networks quickly, but it also gives you a graphical representation of how the model was performing and allows you to exchange that information with one another.