Regardless of its end goal, any software project must go through some common set of steps from ideation to deployment. For example, data science projects, in general, are software projects, and so they need to go through the same development process. This development process contains steps such as ideation and planning, design solution, implementation, testing the software, deploying it, and maintaining it. Although these steps may vary depending on the actual project you're building, you will go through these steps in some form in the majority of the time. Today's article aims to discuss the last steps of a data science project, especially the project testing and maintenance.

The rise of deep learning is accompanied by ever-increasing model complexity, larger datasets, and longer training times for models. When working on novel concepts, researchers often need to understand why training metrics are trending the way they are. So far, the available tools for machine learning training have focused on a "what you see is what you log" approach. As logging is relatively expensive, researchers and engineers tend to avoid it and rely on a few signals to guesstimate the cause of the patterns they see. At Microsoft Research, we've been asking important questions surrounding this very challenge: What if we could dramatically reduce the cost of getting more information about the state of the system?

TensorFlow becomes the de facto standard for creating machine learning models. Find out which tools you can use to get the most out of the framework. TensorFlow is an open source machine learning framework. It is basically a library for numerical calculations based on data stream graphs. The graph nodes represent mathematical operations, while the edges of the graph represent multidimensional data arrays (tensors) flowing between them.

Debugging is one of the most difficult aspects in the lifecycle of deep learning problems. The recent advancements in deep learning frameworks have lowered the entry point for creating really sophisticated models that are both effective and hard to interpret at the same time. Very often, researchers need to understand why the metrics of a specific model are trending in certain direction and they rely on relatively subjective techniques to do so. Additionally, the ecosystem of debugging and visualization tools hasn't evolved at the same speed of the development stacks so very often engineers end up creating models that are next to impossible to debug. Recently, Microsoft Research open sourced TensorWatch, a new tools that takes a new approach to solve the debugging and visualization of deep learning programs.

The rise of deep learning is accompanied by ever-increasing model complexity, larger datasets, and longer training times for models. When working on novel concepts, researchers often need to understand why training metrics are trending the way they are. So far, the available tools for machine learning training have focused on a "what you see is what you log" approach. As logging is relatively expensive, researchers and engineers tend to avoid it and rely on a few signals to guesstimate the cause of the patterns they see. At Microsoft Research, we've been asking important questions surrounding this very challenge: What if we could dramatically reduce the cost of getting more information about the state of the system?