Become a member
Sign in
Aakash Sabharwal
Aakash Sabharwal

Aakash Sabharwal

70 Following
54 Followers
  • Profile

  • Claps

  • Highlights

Highlighted by Aakash Sabharwal

See more

From Ranking Tweets with TensorFlow by TensorFlow

By tuning the parameters of the warm started model, we were able to also see a significant quality gain (0.4% reduction in relative log loss) compared to cold start.

From Ranking Tweets with TensorFlow by TensorFlow

Warm starting from pre-trained model weights can achieve the same or better model quality on new data, with just 20% of the training examples compared to a “cold start” model.

From Ranking Tweets with TensorFlow by TensorFlow

…d not guarantee that the same feature value would be mapped to the same bin index on repeated runs. Fortunately, the modular nature of TensorFlow, being a programmatic Python based platform, made it easy to swap out the discretizer implementation used with a tweaked version of the implementation that does guarantee preserving bin identifiers across multiple runs of the discretizer. We were able to achieve the swap with a 4 line Python snippet that looks like this:

Claps from Aakash Sabharwal

See more

The Bullish Case for Bitcoin

Vijay Boyapati

L1 and L2 Regularization Methods

Anuja Nagpal

How Data Science Helps Power Worldwide Delivery of Netflix Content

Netflix Technology Blog