30DaysML: Day25

I started with trying to understand this implementation of RNN, which involves some language modelling. Soon I had to go back to understand basics!

So I watched few hours of Oxford lessons here, more precisely the Lecture 3 and 4. Way better introduction to language modelling than other tutorials I've seen so far.

Must not lose focus! 


30DaysML: Day24

I went back to TensorFlow tutorials. Since I still don't get high level APIs, I figured best way is to understand these smaller programs and play around with them. 

Also, got completely distracted by some climate change reports, then spent time day dreaming about end of the world. Not productive, Harsha.

30dayML: Day20

Spent quite a bit of time re-writing / adjusting my cat-or-dog CNN network. Now I have an easy way to feed images to the model. Accuracy is still less than 50% though. Hyper-parameter tuning, here I come!

On a side note: Now I see why I should have GPUs. Each run takes 30 to 60 minutes. 

Once I do get to some acceptable accuracy (>90%), I'll move on to RNNs and Voice recognition. Although I still don't know much about Tensorflow high level APIs. OR should I learn Keras? Uggh.

Mood: Sherlock Holmes