Yes we did it!
We’ve just successfully tested an almost seamless integration of the most popular Machine Learning framework, Tensorflow, into the most intuitive and yet powerful business analytics tool, Microsoft Power BI by developing an extremely easy to use Power BI Custom Visual thus opening to unlimited possibilities in terms of advanced zero coding data analysis and modeling in Power BI.
TensorFlow is a software library for machine learning and artificial intelligence developed and mantained by Google. It can be used across a range of tasks, but is used mainly for training and inference of neural networks. Tensorflow is one of the most popular high performance deep learning frameworks, alongside with PyTorch.
Our first experiment consists of a regression engine or “modeler” for the identification of descriptive and predictive models though multiple linear and non-linear regressions. Whatch the demo video here:
In this demo, the Tensorflow Regressor Custom Visual asynchronously trains three different models:
- A simple Multiple Linear Regression Model made of 1 Layer and 1 Neuron;
- A Neural Network Model made of 2 Layers (1 hidden with 100 Neurons);
- A Neural Network Model made of 3 Layers (2 hidden with 50 Neuron each).
A similar interactive demo is available online at: DATITALIA – CPS at report page 7.
The next step will be the development of a Power Bi Custom Visual that allows the interactive configuration of a custom neural network for multiple regression modeling with Tensorflow in Power BI as well as model saving and reloading without writing a single line of code.
After having set up the kind of model, users will be able to experiment each model by simply dragging Power BI columns in the features and outcome fields on the Visualizations Panel, adjust the number of epochs and save the model for further analysis on incoming data.
Refer to analytics@cpsweb.it for more details and requests.