Write Up

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Hi.I am Venkat.

I still consider myself a rookie when it comes to Neural networks. I am well versed in linear
regression learnt all the basic maths about them.In the past I have implements fron
Convolutional networks and one Reinforcement learning network using Ray rllib and also one
stock value forecast using LSTM(recently with one of my friend).

Seeing this problem, at first I thought of implementing some reinforcement learning network, but
setting the reward function and kind of things looked tough. So I surfed through internet for some
ideas ,I got both LSTM and Reinforcement learning as suggestions.As I was busy with quizzes and
DAs I started focussing on it only from 21st April Night. I reffered one LST network notebook in
google which was using some module called cntk which is preinstalled only on Azure VMs. So, I
decided to build a similar network using tensorflow keras, in this kind of short time I got some
base code from chatgpt, and the debugging processs took a whole lot of time.I had to rewrite
much of the logic and the shapes of arrays and tensors are so messed up in chatgpt response. I
had to debug much things manually. Doing this project I learnt about how to write a custom epoch
and different ways in which one can use fit function and what other applications are there for
LSTM.My code now has manny comments written by both me (and also chatgpt).

I completed all debugging process by today evening. So, unfortunately I can’t even complete one
epcoh of training its taking more than 2 hrs for ½ epoch (vs code crashed twice during training).

Unfortunately, in this month start I uninstalled cuda and everything got versions changed for using
GPU in pytorch as I worked with my friend in stock market prediction(using pytorch).The versions
are not compatible with my current tensorflow installation.In short, the training process is still
not done with even one epoch.

You might also like