NOTE: This StatQuest was supported by these awesome people who support StatQuest at the Double BAM level: S. Kundapurkar, JWC, B. Bellman, BufferUnderrun, S. Jeffcoat, S. Handschuh, J. Le, D. Greene, D. Schioberg, Magpie, Z. Rosenberg, J. N., H-M Chang, F. Pedemonte, S. Song US, A. Tolkachev, L. Cisterna, J. Alexander, J. Varghese, K. Manickam
Awesome! Thanks for the nice and clear explanation.
this is very Awesome. I love it <3.
Hi this is absolutely golden for me, also if I’ll have to implement these from scratch for a thing I’m after.
I’m a newb at ML (I can understand the terminology and stuff after years of casually reading about it).
While working on an innovative/new type of game I have invented a new fuzzy search algorithm which could theoretically replace CTC or maybe even do more fancy stuff.
I’d like to test it out as a loss function for an LSTM and then run some benchmarks.
I’m currently stuck with it in Swift (using Accelerate.vDSP on M1 CPUs) in a very advanced state (it even comes out with 0 to 1 closeness values and lots of possible matches) – prolly I won’t even have to port it to Python (needs a lot of complicated initialization, its a lot of code) and a local machine RPC would suffice for the testing.
Could you point me in the right direction? Like to working LSTM exmaple with CTC for NLP/chat bot (not speech or handwriting as it would be difficult to set up I think) and a freely available dataset?
Or a seq2seq translation maybe?
I’m a newb at these so please excuse inconsistencies
Unfortunately I’m currently working on PyTorch + Lightning implementation. 🙁