What do you like best?
The ease of building deep learning models and the high level API's. One can build all kind of architectures pertaining to deep learning. One can use it to build models to solve computer vision problems, perform speech analytics, text analytics, seq2seq models. It supports major algorithms such as ConvNets, RNN, LSTMs, Seq2Seq, It also includes some of the pretrained models which can be customized and trained with new datasets. It can also scale to use multi -gpu systems out of the box without much configuration.
What do you dislike?
Some of the benchmark results show it doesn't train fast, I hope the team is working on making it faster. Also it doesn't include other ML models for comparison.
Recommendations to others considering the product
What business problems are you solving with the product? What benefits have you realized?
We are trying to use multi-layer neural network in customer analytics space. We use various models and TensorFlow is easy to implement using high level api.
Personally, I have used it in building forecasting models, sentiment analysis, text analytics, language translation, general adversarial networks. The implementation for all the models were very easy and it provided great benefits in terms of using multi-gpu systems efficiently.