• AI Black Belt Blog

    The 5 Best Python Libraries for AI

    • TensorFlow: TensorFlow is Google’s open source framework and probably one of the most famous and arguably one of the most powerful frameworks for the AI development it can also be used with other libraries such as Keras which is going to be explained below. TensorFlow provides TensorBoard for visualization. It lets you build more complex neural architectures. It also has some disadvantage which includes there are no pre-trained models, and it is not fully open source.
     
    • Microsoft CNTK: Microsoft CNTK  which stands for computational network toolkit is Microsoft's reply to the google’s tensorflow and let me tell you Microsoft has done a pretty good job. It is mostly used where the hardware required for the AI system is larger cause that is where we will have to use many servers and it works on the servers better than the tensorflow. The CNTK almost catches the tensorflow especially in the speed department but lacks in the department of visualization as there is no support for visualization in the CNTK. It can support Java, C++, and C# too alongside with python.
     
    • Theano: Theano is a strong competitor to tensorflow and already giving it run for its money. It is used where multi-dimensional data is present as it handles that kind of data efficiently. Its clear use of GPU gives an advantage to the platform. That is the reason it is one of the most used library in the companies who deploy large scale computationally intensive operations. It has its own flow which includes its buggy performance on the AWS platform.
     
    • Keras: It is an API structured libraries unlike tensorflow and CNTK that is used to create a convolution network of 2 to 3 layers to any general recurrent network. It comes with prepackaged network and datasets which makes it a favorable platform where you need to save time for testing the network but it is not all good. It is not so good with the recurrent networks also it is hard to make new architecture other than that already come with the library are hard to design.
     
    • Sci-kit Learn: Sci-kit Learn is a made with other Python libraries such as Scipy numpy and matplotlib it should be used for the statical modeling which includes the classification, regression, and clustering. It comes preloaded with most of the Machine learning algorithm which helps in the fast development. It is not good for making machine learning models and also it is not good with using the GPU’s efficiently.
    Add paragraph text here.