Day 8
Today morning I had my Soft Computing examination. It was easy - only numerical problems related to fuzzy sets, relations etc.. I got briefed about my internship project. I have to complete 3 tasks within 3 months.
- Air Demand Analysis - covert a MATLAB model to a python ML model
- Reciprocating Compressor - predict component failure
- Power BI customisation - there's no desktop app for Mac :( I'll have to try installing it in a Virtual Machine
Day 9
Online classes started after a peaceful Christmas week. But its still a 1000x times better than offline classes.
I'm working on the Air Demand Analysis. I wanted to finish it today - but unfortunately things don't always go as per plan. I'm stuck at making LSTM Networks in Keras. I'll have to learn it in detail tomorrow to understand what's happening. I started doing my 10 days overdue soft computing assignment on Backpropagation networks - I didn't pay attention to the google classroom notification since it was Christmas hols.
Day 10
Submitted my way overdue backpropogation assignment. I found an awesome poster on GitHub which explains visually the data flow which happens in LSTM implemented using Keras.
I had my first interview with MLH for their Externship program. I really hope I can make it to one of their cohorts in 2021. I would love to work with other students on interesting tech. The interviewer was really friendly and I loved listening to her experiences in tech.
Finally removed all errors in the ML model. But guess what? Next bomb drops - Model doesn't learn.π₯Ίπ₯Ίπ
Day 11
Happy New Year! A new beginning. I hope I can stay committed to my goals and improve myself a tiny bit every single day in 2021.
I started the year by solving another Backtracking problem - Rat in a Maze. I also did the #Leetcode daily challenge problem. It was about fitting small array pieces to form a single array.
I'm back at LSTMs and I understood that I'm making many fundamental flaws. I'm just so impatient about finishing the project!
Day 12
I implemented a crossword problem using backtracking. Accessing elements outside matrix will definitely lead to segmentation fault. The words to be filled in crossword were given by a semicolon separated list. Here's a clean piece of code to accept such inputs:
s="CALIFORNIA;NIGERIA;CANADA;TELAVIV"
stringstream ss(s);
vector<string> result;
while(ss.good()){
string substr;
getline(ss,substr,';');
result.push_back(substr);
}
Microprocessors assignment due tomorrow. I wrote a few 8051 programs. After 4 days of intense LSTM research, my models finally learns something even though there is a huge error. But I'm happy that it works at least syntactically. I need to work on:
- What's LeakyReLU? Is it needed.
- Have I factored in all the MATLAB training options into my model?
Day 13
Participated in my first ever Leetcode contest(Weekly Contest 222). Bad rank tbh 5810 / 6575. No..I thought I'd be able to solve at least 2/4 questions...ughh...Next time I'll do better.
I did 2 more application questions on sorting and searching. One question was really good - I had to look at hints to understand it was a binary search application. It was a relaxing Sunday.
Day 14
Wrote an analysis for yesterday's Leetcode contest. 2 more to do. Read a bit more of James Clear's Atomic Habits. Focus on systems rather than goals.
Finished off yesterday's Leetcode challenge - did backtracking for permutations. I have to find a better solution though but I'm happy I did it by myself. I completed today's challenge too - it was pretty simple - merging 2 sorted Linked Lists.
I started learning bitwise operators. They are a really powerful tool. I submitted my weekly LSTM progress - not a good model to be honest.
Take care,
@stratospher
Top comments (0)