Machine Learning- Data Science

I was feeling confused & overwhelmed in the RNN's learning process :
So I wrote the RNN world learning Evolution in short to map my learning process

This cleared my 100% confusion.
Now next is "Attention is all you Need" https://t.co/LZiMTeaNrc

Day 254 : DataScience Journey
Using industrial electricity production data, optimising probabilistic time-series forecasting model and then applied regime-aware SHAP to answer a practical question: does the model think the same way during different economic phases? SHAP was used https://t.co/LNDLIla21L

Day 253 : DataScience Journey
A single forecast looks confident.
But confidence without uncertainty is usually misleading.
I stopped trusting models that give one number
and started building ones that explain how wrong they might be,Instead of point forecasts, I modeled https://t.co/Auo5huDT2M

The work of 28 & 29 Jan:
- LSTM's revision
- In depth GRU's learning
- Deep RNN's
- Bi-Directional RNN's
Next some sentiment analysis on Quora Duplicate Questions Dataset

Eager to start with Transformers ! https://t.co/CZrXQ6BLR4

Day 247 : DataScience Journey
Started with a new dataset today; monthly electricity production and spent time just understanding the data before jumping into models. seasonality plt across yrs made things very clear. The shape repeats every year, but the amplitude isn’t constant. https://t.co/9zszV9VBc6

Introducing Ramartha —

> a computational framework for semantic and narrative analysis of Shri Ramcharitmanas
> It models meaning at the verse and narrative level
> enabling concept-driven exploration in a low-resource NLP setting

> build live app:
https://t.co/p8hhhb4ydZ https://t.co/kAeySjAJg9

Day 246 : DataScience Journey
This phase really connected the dots for me. After building direct multi-step forecasts with XGBoost, I added prediction intervals using residual-based uncertainty. Technically it’s just estimating error spread, but the plots made the idea of https://t.co/fibFKGezma

Day 244 : DataScience Journey
Checked XGBoost feature importance today and the model made its priorities very clear lag-12 is doing most of the heavy lifting, which makes sense because this data is all about yearly seasonality. lag-1 still matters (recent momentum), and the trend https://t.co/REqCm7bRtt

Day 241 : DataScience Journey
Not gonna lie, time series is slowly teaching me humility.
I thought adding “better features” would instantly fix the forecast, but it doesn’t work like that, compared a basic feature set (lags + rolling stats + month) against an improved one where I https://t.co/UXh74DiSIe

Want projects that make your portfolio stand out? Check this out:https://t.co/51XQi6Kdis https://t.co/x39rU1r192

Coding till 24:00 ;) https://t.co/XpPjzOtkhb

Working on A Low-Resource NLP Project - Linguistic Analysis on ShriRamcharitmanas
> Using Sanskrit/Awadhi language
> Training the Word2Vec & FastText models
> Comparing both models on Word level & Verse level Semantics
> Eager to complete this project and share it with you all :) https://t.co/OoyTV7tRjq

need some honest feedback on my resume, roast away and tell me what to fix to get a ml internship https://t.co/Mb5wXLaEvJ

Day 236 : DataScience Journey
Today’s work on the Telco Churn dataset. The first real hurdle was the data itself, TotalCharges looked numeric but had blank values stored like text, so converting it and dropping invalid rows was necessary before anything meaningful could happen https://t.co/JXpci7ndSe

Day 235: DataScience Journey
A real fraud detection system isn’t about building the “best model”… it’s about catching suspicious
transactions without flagging everything and wasting manual review.
That’s what makes it tricky although are rare, but missing one can be expensive. https://t.co/lQphjhDfDb

Day 234: Data Science Journey
Logistic Reg. is one of those topics that looks basic on paper, but hits different when you actually run it and see the boundary forming. The main thing that stood out was how it’s not just predicting 0/1, it’s giving a prob. like “how sure am I?”. https://t.co/9cfEf91HJG

Day 232 : Data Science Journey
Today’s session felt genuinely productive, practiced a bunch of SQL ques on aggregate functions and grouping (like Average Selling Price, etc.) and it’s slowly starting to feel less like “writing queries” & more like pulling real insights.
Then https://t.co/T2WvfhfVlJ

Day 231 : Data Science Journey
Today, did A/B testing the right way, used the Bank Marketing dataset and treated it like a real product experiment: created a conversion metric (subscribed = 1), split users into Control (A) and Variant (B), and then actually verified that the https://t.co/7rKOrZdjB7

Just dropped a deep dive blog on LSTM

Why vanilla RNNs break, how gates control memory, and how stable sequence learning actually works —intuition → math → code.

If LSTM ever felt like magic, this blog reverse-engineers it.

- Give it a read of 5 mins

https://t.co/gegoHn9WoC https://t.co/fcw8ldCyEt

After a long time, I finally wrote a new blog

Why 97% Is Not Enough Building a high-sensitivity pneumonia screening system with DenseNet & Grad-CAM

• ~10 min read
• Real project learnings & why safety>accuracy ?
• give it a read 🙂

Enjoy reading : )

https://t.co/rmB81tAPjX https://t.co/YViOL2EZmv