Keras, the right entry to the Deep Learning world
May 19, 2021
Keras, the right entry to the Deep Learning world
Score 8 out of 10
Vetted Review
Verified User
Overall Satisfaction with Keras
Keras is being used together with TensorFlow, in our new Data & AIML department. Many business use cases are implemented on top of it. We use Keras as high level API, responsible for model designing, model training, model evaluation and model inference/prediction. It is part of our final Model As Service product, deployed in our production environment, allowing various business application [to] consume its prediction output and take the important decision as early as possible.
- As the high level API, clean and neat, allowing the data scientists to develop the standard deep learning model very quickly (using the mature and existed algorithm module blocks)
- Seamlessly integrate with [a] couple of Deep Learning framework, although we only use TensorFlow, and since TF2.0 Keras has become the 1st class native API
- Keras model itself is not thread safe, and the documentation regarding how the multi thread works during the prediction time is not quite clear and enough, we still need [to] tweak the TensorFlow old (session + graph) way to support our Model As Service, in the high concurrent scenario, but with limited memory constraint.
- Some API and default implementation still has a lot [of] improvement room, for example, the checkpoint callback, only saves the "best" model, natively can not save all other metadata of the "best" model, we currently have to extend that by ourselves to meet our project requirement.
- Some advanced topic, like distributed training, documentation is not so clear and very hard to find the real code example.
- Developer friendly descriptive API
- Natively support TensorFlow 2.x
- Keras enables the prediction needs based on patterns hidden in the big historical data, those predictions can guide the business work more efficiently, reducing a lot of operation cost while increasing the accuracy of the service.
- Keras enables a lot [of] automation in the business flow, those jobs previously required heavy manual work.
- TensorFlow and Pytorch
As Keras is the high level API, so using Keras, we don't have to be bothered by the low level TensorFlow complexity, and we can reduce a lot coding and testing efforts.
Do you think Keras delivers good value for the price?
Yes
Are you happy with Keras's feature set?
Yes
Did Keras live up to sales and marketing promises?
Yes
Did implementation of Keras go as expected?
Yes
Would you buy Keras again?
Yes