Day 43

I'm a little stressed about tomorrow - I promised to build a preliminary model. My model takes too much time to train and follows a data driven approach. Machine Learning isn't as simple as I thought 😂 - there's an intuition which can be developed only with practise. I read up more research papers here.

I also did a few easy leetcode questions:

1. Floyd's Cycle Detection Algorithm
2. Right view of a binary tree

Day 44

Yay and Ugh :( Visit got postponed to the day after. I didn't finish making the model. I made 1 more power BI visual which will help display the excel tables as it is. We can select particular rows and columns to display as we choose. Semester exams looming in the horizon.......How will I manage exams and ML models?😂

Day 45

Started learning microcontrollers for semester exam. We have to learn about the 8051 microcontroller. Microcontrollers are single purpose devices with RAM,CPU,ROM,I/O and timer all on a single chip. There are soo many bock diagrams to learn by heart:(

I also made another visual using Tableau which shows a block diagram relationship of various entities.

Day 46

Went to demo the model which didn't finish training. They are really nice, friendly and supportive people and gave me suggestions to improve my model.

1. Correct the window_length and other parameters to consider data for past 2 weeks
2. Make a new feature that will give the probability of equipment failure and test. This seems like a good suggestion. I wonder how we do that though? I have to get it done by day after.

The oil refinery is like science fiction. Everything's so beautifully constructed in perfect geometrical equilibrium with so much metal everywhere. It's very unlike what I imagined.

Day 47

Learnt more about 8051 microcontroller. Checked out cumulative and exponential moving average. Agreed to take a session on a few coding problems...wait..did I just agree to do that? Meh! It cant be that hard.

Day 48

Did more work on exponential moving averages. I selected 4 features and decided to take the percentage change of the features...I need to make a probability formula that would predict the failure.

Day 49

I couldn't make a probability formula which accurately captures the relationship for failure. I'm exploring other options now. I came across an amazing python library sktime designed especially for working with time series. I loved it.

One major difference is the way you pass input into the functions of this library. Normally we pass dataframes as inputs. Here each row is the Dataframe is an array.

     X
[,,,,,] 
[,,,,,]
[,,,,,]

Until next time,
Bye!
@stratospher