Day 36
Wrapped up all submissions for my Design project course.
Working with Unity was a good experience. The only problem was the high GPU requirement lol.I scraped through though.
Brackeys is an amazing channel for learning Unity!Turing machine quiz this week - so learnt that.
Day 37
System Software Lab Exam happened. We had to implement a 2 pass macroprocessor with unique label generation and also round robin process.
I did some more of Datacamp's Time Series course. I read a tweet about the need to include more factors into the Stock Market's Time Series Model for real world prediction. I wonder if it's the same for ADA prediction?
I listened to an absolutely amazing Lex Fridman Podcast on AI and Physics with Max Tegmark. It was an eye opener for me - AI's biggest threat to humanity is that we don't properly understand the neural networks and a lot of technology trusts user privacy and all on the neural networks. They discussed about how the physics pattern of thinking can help counter these as it focuses more on ofai(old fashioned ai), discovering new laws of physics, Aliens and so much more.
Day 38
I learnt how to use Tableau for getting data insights. It's great for creating visualisations - I love the story feature which lets you create a hybrid between powerpoint and dashboards. But working with Microsoft Power BI feels more intuitive for me. It has more features for formatting the data and getting it to the whatever form you want.
And I only completed 56% of January Leetcode challenge - last week took it's toll on me. But yeah - It's way better than doing no Leetcode. I solved 4 questions with concepts involving greedy approach, mathematical observations and bit manipulation. I found the C++ STL for sets really awesome.
Day 39
More data analytics and visualisation. I created a Tableau story and Power BI dashboard for the same data set. Reached 'Man who didn't look right' in 'Atomic Habits'. The part that stuck with me is:
"Until you make the unconscious conscious, it will direct your life and you will call it fate" - Carl Jung
Started leetcode Feb challenge. Learnt basics of greedy technique.
Day 40
Started working on next ML use case - it's for forecasting equipment failure. Unfortunately, it's real world data set and not a toy data set. They don't have a column which we should predict like in Kaggle. Instead they have given us the maintenance logs and we have to infer from there? We have to do lot of data cleaning too.
Day 41
I spent the day researching for similar use cases. I think this comes pretty close to what I want:
https://github.com/Azure/lstms_for_predictive_maintenance
I read a lot of research papers too about predictive maintenance and it's pretty fascinating. My biggest challenge right now is :
- Data cleaning
- Making the target column
Day 42
Found some really cool python functions to make data into the form I want:
# if u r reading from many csv files
li=[]
li.append(df)
li.append(df)
frame = pd.concat(li, axis=1)
# if Date is in this form 01-JAN-15 00:01:00.0 and we need to convert to python date time
frame['Date']=frame['Date'].str[0:7].astype(str)+'20'+ frame['Date'].str[7:].astype(str)
frame['Date'] = pd.to_datetime(frame['Date'], format='%d-%b-%Y %H:%M:%S.%f')
# for setting a value in pandas data frame when a condition is satisfied
frame.loc[(frame['Date'] >= dt.datetime(year=2015,month=1,day=1)) & (frame['Date'] < dt.datetime(year=2015,month=1,day=2)), 'is_failure'] = 1
Bye!
@stratospher
Top comments (0)