Published inBetter ML·PinnedUsing Transformer Model for storing knowledge and Question AnsweringCustom Training Huggingface GPT2 for Closed Book Question Answering — Introduction The Hugging Face GPT-2 model is a state-of-the-art language model that can generate natural language text. It is built on the Transformer architecture, which allows it to learn contextual relationships between words and phrases in a text. Contextual relationship means that it is able to give a different embedding…Transformers18 min readTransformers18 min read
Published inBetter ML·PinnedImage pre-processing for TF Serving via OpenCV, Pillow, TensorFlow tf.image.decode* , Keras-Retinanet ExampleExample Input Images (all images are attributed to their owners, just showing here for technical demo)Image Processing10 min readImage Processing10 min read
Published inBetter Software·PinnedOptimizing any TensorFlow model using TensorFlow Transform Tools and using TensorRTModel Optimization and reducing precision from FP32 to FP 16 for speedup and reducing graph size. — What is this article about? Whom is this meant for? More than an article, this is basically how to, on optimizing a Tensorflow model, using TF Graph transformation tools and NVIDIA Tensor RT. This is a bit of a Heavy Reading and meant for Data Science Engineers than Data Scientist. Also, more than a how-to, it also illustrates the bugs…TensorFlow19 min readTensorFlow19 min read
PinnedBackpropagation Deep DiveBack Propagation with Softmax output and CrossEntropy Loss — Update — Please see the updated one at https://alexcpn.github.io/html/NN/ml/8_backpropogation_full/ The internet is teeming with articles explaining Backpropagation. Why one more? Well, I set out to understand this and thought I understood this, having derived out one and done a toy example too. But then I tried to code a simple…Backpropagation9 min readBackpropagation9 min read
Published inBetter Software·PinnedFrom Data Warehouse to Data Lake to Data LakehouseWhat’s for what, what you need, and what are the advantages and limitations — Before we go to Data Lake we need to go through the other Data Store technologies, to see the full picture and to understand, what has changed and what has not. On the journey, we will build a Data Lakehouse using the current state of art popular stacks and check…Data Science14 min readData Science14 min read
Published inBetter Eating·1 day agoTo Egg or Not to EggThe controversy surrounding dietary cholesterol to blood cholesterol — Short answer. If you are growing stage — say till 25, one or two eggs per day are good. After that one or two eggs per week is better; and maybe egg whites daily if you lack other protein sources. On average most people all over the world die due…Eggs7 min readEggs7 min read
Published inBetter Investing·May 21The Price To Earning P/E RatioUnderstanding the Pros and Cons of using P/E Value for Investing — “Beware the Jabberwock, my son! The jaws that bite, the claws that catch! Beware the Jubjub bird, and shun The frumious Bandersnatch!” Jabberwocky by Lewis Carroll Investing in stocks is often seen as a pathway to building wealth, and the Price-to-Earnings (P/E) ratio has long been used as a beacon…Investing15 min readInvesting15 min read
Published inBetter Investing·Apr 18Infosys has had one of the worst Quarters and I am still longThe talk of the day is the Infosys results and the subsequent stock market tank of around 10 per cent in a single trading day. This very fact goes on to prove that many people or a vast majority are not investors but have a trader mindset. There is nothing…Infosys4 min readInfosys4 min read
Published inalexpunnen·Mar 29The Elephant Whisperers and WorkWhat this documentary tells about work and life — I watched this recent Oscar Award-winning documentary. Coming from being indulged in very immersive and technically difficult documentaries from National Geographic and BBC, this rather simple documentary disarmed me for a moment and then embraced me in the simplicity of it all. At the heart of this are not elephants…Philosophy2 min readPhilosophy2 min read
Published inTechlogs·Jan 16Grafana Loki with AWS S3 backend through IRSA in AWS Kubernetes ClusterUsing IAM Role based Access for Service Account — Why Loki? Primarily because ElasticSearch (Not Opensource anymore)/ OpenSearch (Open Source) uses Block storage/EBS in AWS and that is pretty costly compared to S3 which Loki uses with BoltDB # Need to do #1 Check Read performance from S3. The write would be batched and should be fine I guess…Loki5 min readLoki5 min read
