This post is aimed to summarise what I learned from the
fastbook reading session with Aman Arora from Weights & Biases. It is a study group which has come into effect thanks to the collective effort of Aman and the organization Weights and Biases for which I am really grateful. The entire session is recorded and can be viewed below
This post summarises the contents of week 3 with the help of a personal project that I did to replicate the concepts taught in this session. I hope you find this helpful :)
Deep Learning/ Machine Learning, unlike traditional…
In this post, I wish to summarize what I learned from the fastbook session on the reading of first portion of the fourth chapter of fastbook i.e. MNIST Basics. It was a really fun session conducted by Aman Arora from Weights & Biases. Parul Pandey shared some really nice advise on how blogging as an activity can fuel ones curiosity and overall help to build a good Data Science career as well.
The session is available for streaming on YouTube and below’s the link for the same.
In this age of Data, python has become the most widely adopted language by many developers around the globe. This means that there are many potential libraries sitting there waiting to be utilized.
While python has recently gained the ethos for being the first choice to analyze data and has a very wide spectrum of applications spanning low to heavy usage for many tasks; Microsoft Excel has a history of, and in most cases still is an inevitable tool used to analyze/munge/visualize data.
Imagine if we could utilise the open-source libraries which are available in python for automating our workflows…
I have had been meaning to study an area of deep learning called metric learning and I figured out a great learning use-case that piqued my interest and made me delve straight into it. It’s the face recognition problem.
But wait, there’s a slight twist here. We’re not doing plain human facial recognition because it’s just meh, very commonplace!. As a proponent of the otaku culture, I thought let’s do facial recognition with anime characters!!
This would be a three post series which would talk about every aspect of the project right from problem definition to model evaluation and deployment…
This post is aimed toward demonstrating the use of session state in
Streamlitfor storing info about certain variables and prevent them from updating across runs. But what is
Streamlit is a python library which is aimed to make the process of building web applications to showcase your work very easy for python developers. Until the arrival of this package
Djangowere the goto libraries which developers chose to use in order to develop and deploy their application over the web; however both these frameworks required the user to write HTML/CSS code to render their work as a…
This is the final post in a three part series which demonstrates the process of building a simple ML model and deploying it to an API endpoint. Previously, we have seen how to scrape data from the web for building a dataset of dialogues narrated by the two characters Captain America and Iron Man. Subsequently we saw an expository overview of the Naive Bayes Algorithm and built a model (dictionary of conditional probabilities) to identify the author given a dialogue. You can refer this post for getting an overview of the model building phase.
As an ML practitioner, it is…
This post is the second in a series of three posts which aim to build a naive bayes classification algorithm in an end to end fashion. In the previous post, we looked at curating the data for this problem: We scraped dialogues for both Captain America and Iron Man from the transcripts of Captain America: The First Avenger, Iron Man and Avengers: Endgame respectively. We then processed them to remove all action/scene description cues and only retain the spoken word by both our characters and created a dataframe out of the same and saved them as csv files.
This post is the first in a three part series which aims to implement a Machine Learning Algorithm in an end to end manner using dockers. The underlying objective of this problem is to build a classifier using Naive Bayes Algorithm which could tell if a sentence is more likely to have been uttered by Steve Rogers aka Captain America or Tony Stark aka Iron Man. I will be using both the names of both characters to suit the context through the remainder of this post and subsequent ones.
Note: Some websites do not allow for scraping of data. Please…
With the surge in use of video calling services during the COVID lockdown, many players are offering a service where the user of the service could blur the background or add a custom background etc. (Zoom, MS-Teams etc.) This is a classic use case of image segmentation where the object of interest is located and the pixels barring this region are modified/substituted.
There are many deep learning architectures which could be used to solve the instance segmentation problem and today we’re going to useDeeplab-v3 which is a State of the Art semantic image segmentation model which comes in many flavors…
The table of contents of this post is as follows, click on any bullet to toggle to the respective section
Time series data has been around for quite a while and many people have studied them and came up with ways to analyze the data to find meaningful insights and subsequently use it to make reliable predictions as to how the series will progress in the future. Stock market…
AI Developer, Data Science Enthusiast