Daily log to track my progress on the 100 days of ML code challenge.
- Started with the intro to machine learning course on Udacity
- Learnt the basics of a Naive Bayes classifier on the
iris
dataset - Working on classifying the
Stanley Terrain
dataset and graph the decision surface
- Working on the Naive Bayes mini-project to classify email.
- Tried really hard to make the python 2.7 code compatible with 3.6 and learnt about
dos2unix
and pickling of data. - Completed the Naive Bayes project with accuracy of 90.24% (Need to improve it!)
- Improved efficiency to 97.869% and completed the mini-project.
- Started the lesson on Support Vector Machines.
- Completed Week 1 of Mathematics for Machine Learning: Linear Algebra, a course from Imperial College London on Coursera.
- Completed the SVM mini-project with 99.08% accuracy using an
rbf
kernel - Started the lesson on Decision Trees
- Working on the Decision tree mini-project
- Referred to 3Blue1Brown's Essence of Calculus playlist
- Completed the Decision Tree mini-project
- Learnt about the K-Nearest Neighbours classifier and implemented the same
- Implemented the KNN classisier after referring to this Medium article
- Watched 2 more videos from 3Blue1Brown's Essence of Calculus playlist
- Watched Siraj Raval's video on classifiers
- Completed the lesson on datasets and questions to gain key inferences
- Completed the lesson
WEB 3.0
from Siraj Raval's Decentralized Applications playlist - Implemented the RandomForest classifier and read up about adaBoost
- Completed the lesson on Regressions and implemented the same in the mini-project
- Completed the analysis of outliers in the enron dataset and the Q&A on the analysis
- Completed the lesson on unsupervised learning (K-Means clustering)
- Implemented K Means clustering on the Enron dataset
- Completed the lesson on feature scaling (MinMaxScaler)
- Stemming using NLTK(Natural Language Toolkit)
- Completed the lesson on text learning
- Completed implementing the string processing techniques in the dataset (17578 emails)
- Completed the lesson on feature selection
- Implemented Lasso regression to understand regularization
- Completed the lesson on dimensionality reduction
- Working on the eigenfaces mini-project
- Completed the lesson on Validation and its exercises
- Completed the lesson on evaluation metrics and its exercises
- Started the Deeplearning.ai course Neural Networks and Deep learning by Andrew NG
- Completed the intro to machine learning course on Udacity!!
- Working on finding the persons of interest from the Enron emails dataset
- Completed Week 1 of the Neural networks and deep learning course
- Read up about Tensorflow from the documentation and medium articles
- Watched 2 Coding Train videos to understand Tensorflow.js
- Implemented classifier and regressor using tensorflow and compared the same with the sklearn implementations
- Learnt about the softmax, one-hot encoding and cross-entropy loss minimization using gradient descent
- Learn best practices to handle missing data and effective feature selection
- Practiced the preprocessing workflow
- Built a basic stock predictor app that predicts the value of the stock and the value of the company
- Referred to this video by Siraj Raval
- Analyzed the Mars HI-SEAS dataset using SVM (and PCA) to unearth outliers and analyze for predictive analytics
- Performed data wrangling and analysis using dplyr in R
- Started the Intro to deep learning course by Google Brain's principal scientist Vincent Vanhoucke
- Built a neural network with 84% accuracy for the notMNIST dataset
- Completed lesson 1 of the intro to deep learning course
- Working on week 2 of Andrew NG's course on deep learning and neural networks
- Implemented gradient descent from scratch
- Completed assignment 1 of week 2
- Implemented logistic regression using a neural network approach to classify images
- Completed Week 2 of Andrew NG's course
- Implemented gradient descent form scratch
- Learnt more about activation functions sigmoid, tanh, ReLU and leaky ReLU
- Learnt about the advantages and differences between tensorflow.js and tensorflow
- Completed planar classification assignment
- Completed Week 3 of Andrew NG's Neural Networks course
- Started Week 4 of the course
- Completed all lecture videos of Week 4 pertaining to deep neural networks
- Working on the programming assignments
- Completed assignment 1
- Working on a cat-notCat binary classifier using a deep neural net
- Completed Week 4 of the course and obtained the certificate!
- Learnt the math behind Frobenius norm and regularization
- Started course 2 of Andrew NG's Deeplearning.ai specialization
- Completed week 1 materials and working on the optimization exercises
- Implemented l2-regularization from scratch
- Implemented dropout (forward and back-prop) from scratch
- Implemented Gradient checking from scratch
- Completed Week 1 of the course
- Implemented mini-batch gradient descent with momentum
- Implemented Adam optimization from the ICLR 2015 paper
- Completed week 2 of the course
- Implemented batch normalization from scratch
- Working on the SIGNS dataset to identify numbers from sign language (Using Tensorflow)
- Completed the course on Improving deep neural nets - certificate
- Completed the course on structuring machine learning projects! Certificate
- Learnt more about transfer learning
- Read and practiced from the Tensorflow documentation to better understand the workflow
- Understood the importance of GPUs in Deep Learning and the
tensorflow-gpu
module
- Started Week 1 of Andrew NG's course on Convolutional Neural Networks
- Learnt more about Tensorflow from Jordi Torres' Deep Learning book
- Completed Week 1 of Convolutional Neural Networks
- Learnt about pooling(POOL) and fully connected(FC)
- Attended GDG DevFest 2018! Was a very informative event for ML/AI practitioners
- Working on building a CNN step by step
- Learnt more about Google's ALphaGoZero and why it's such a big breakthrough
- Learnt the very basics of Reinforcement Learning
- Learnt about Basics of RL from David Silver's online course
- Learnt about Pooling layers for CNNs and improved implementation
- Working on Week 2 content of Andrew NG's CNNs course
- Learning about PPOs (Proximal Policy Optimization) in RL
- Learning about rocket launches to build an app to track space-flight schedules
- Building and training a ConvNet in TensorFlow for a classification problem
- Spent some time preparing data from Nasa datasets for the topic "Do YOU Know When the Next Rocket Launch Is?"
- Prepared and pre-processed the data for the Nasa SapceApps competition
- Using the GLOBE dataset to predict effective sunlight cover on solar panels
- Used Monte Carlo simulations and normalization to predict the conversion factor for solar panels
- Used the conversion factor thus obtained to build a calculator to visualize the data
- Worked on CNNs with a 'selu' activation function for better learning rate with normalization
- Revised building CNNs from scratch from Andrew NG's course notes
- Studying for internal exam on the subject of Artificial Intelligence
- Studied for my AI exam on 25th
- This includes pedicate logic, Bayesian statistics, Bayesian networks and partitioned semantic nets
- Gave my AI exam and probably aced it!
- Working on training a model on a scraped data of roof pictures with given dimensions (labelled) into a CNN to estimate the solar irradiance incident on the surface
- Working on a You Only Learn Once model for car detection
- The ML project pipeline is underway
- Went to the Google office for a meetup called #chAI where early stage AI startups explailned the deep learning they have been doing
- fixed all deployment bugs in the Nasa SpaceApps project and hosted the website
- Learnt more about YOLOv2 from medium articles
- Got project guidance and tips on the Solar roof CNNs project from Vibhor Kalra from merak.ai
- He suggested to look into tensorflow.js if browser based real-time models need to be deployed
- Need to learn about deploying a tensorflow project
- Learnt about tf-lite models and their merits and demerits for DL apps
- Improved the prediction model for the solar project and working on the final submission as today is the last day
- Registered for the Microsoft AI challenge to improve Bing's suggestion box answers using DL models
- Re-doing the plan for the next 50 days to get the most done from this challenge
- An implementation from Andrew Trask's blog about sentiment classification to frame problems in deep learning
- Completed CNN implementation from scratch
- Still working on a feedback analysis of the progress thus far to get much more done in the second half of the challenge
- Studying the math behind backpropagation (for CNNs) from Ian Goodfellow's Deep Learning Textbook
- Implemented a fully functional CNN using Tensorflow
- Improved the friend dashboard project
- ALso created a Genomic and AI related github organization for related projects
- Working on GenomicAI's website. Looking to finish it up after Monday's exam
- Revising data science in R from Harvard Prof Rafael's textbook
- Completed half the course by Google Cloud Platform on 'How Google does ML'
- Working on the paper on 'Genomic analysis for persoanlized medicine'
- Earthquakes project using a datalab instance Link
- Project Link
- The Common pitfalls in ML deployment. Gosh it has much more to do with stuff other than ML!
- Completed Google Cloud's first course 'How Google does ML' Link
- Started learning the procedure to prepare and pre-process datasets to bucket in cloud instances
- Working on GenomicAI's about page
- Made UI improvements to the Social Network
- Learnt more about operationalizing ML models for production using the Google Cloud Platform
- Awaiting the scholarship confirmation. I just about finished the content in the course from Coursera
- Working on the site page for GenomicsAI
- Exploring AWS ML APIs as compared to GCP's ML APIs
- Buying parts for my Deep Learning rig. Got a GTX 1080 and an 8Th gen Intel i7 processor. Need to save up and buy the rest of the parts!
- Made some progress on Google's GCP challenge on specializing in ML by 30th November
- Learnt about some of the methods to build and operationalize a recommendation engine
- Working on the prototype for a recommendation engine for user's feed in a social network I am building
- The social network is built on a PostgreSQL database with a flask business logic. Check out my profile for details
- Working on lesson 1 of the content to build a neural network using PyTorch
- Made significant headway in the Social Network project
- Continued the Udacity course on PyTorch
- Started Siraj Raval's Move 37 course for Reinforcement Learning
- Working on the social networking application for my DBMS project. Looking for ideas to include ML concepts in it
- Completed lesson 1 in the PyTorch challenge
- Learnt about the Bellman equation in Reinforcement Learning (Move 37)
- Fixed performance bugs in the 'Social Network' project (which I have to submit soon)
- Working on the PyTorch challenge lesson 2
- Learnt about CUDA programming to utilize a GPU to its max
- Learnt about TFlite models for deep learning on a smartphone
- Learnt about the basics of using Google Cloud along with BigQuery datasets and ML
- Worked on learning about TFX to build end to end Deep Learning models
- Completed the minimum viable project for DBMS lab!
- Completed lesson 3 of the PyTorch challenge
- Working on the DNNs with PyTorch lesson
- Imperative programming in PyTorch and the dynamic front end is more suited for research implementations
- Learnt more about deploying models as low level C++ and the production-ready Tensorflow workflow
- Read about classic CNN architectures like AlexNet, Lenet-5, VGG-16 and Microsoft Research's Resnet
- Learnt the Keras workflow to implement Resnet
- Learnt more about Skip connections with Convolutional as well as ID blocks
- Completed the Resnet implementation
- Working on car/object detection
- Part of the YOLO paper released on June 12th 2015 but without the K-Means clustering for drawing the bounding boxes
- Learnt about implementing non-max suppression and IoU for filtering probabilistic results
- Despite Yolo being a good solution, tried to implement Fast CNNs and Faster CNNs from scratch in Tensorflow
- Poor results on this. It is more or less guessing the solution despite using Adam optimization
- YOLO with K-Means clustering seems like a better option. Will look into it soon
- Learnt about the Computation graph and how paralellizing TF clusters improves performance
- Ordered parts for my Deep Learning PC!
- Have an i7-8700, an NVIDIA GTX 1080 and the MSI A-Pro Z370 motherboard so far!
- Working on Siamese Networks for learning Similarity functions
- Improving Happy House with Face recognition
- Working on Udacity's Lesson 4 of the PyTorch challenge
- Got the O'Reilly Data Science book in R with the Tidyverse
- Working on the problem statement for Hackference Hackathon
- Work on the Hackference hack cancelled as the deadline for documentation submission passed hours before we submitted our proposal!
- Learnt more about GPU and CUDA programming for Deep Learning
- Completed 2 chapters of the O'Reilly Deep Learning with Tensorflow book. It is a fantastic book to read!
- Watched a few videos by SentDex on Youtube for a hands-on refresher in Pandas and Matplotlib
- Preparing for data science internships and thus reviewed Sampling theory and some sample interview questions
- Learnt more about the Tidyverse in R and how statisticians build their workflow in it
- The initial steps include: Exploratory Data Analysis(EDA) with ggplot2, Wrangling with tidyr, dplyr and programming with magrittr and purr
- Had to spend time on exam preparation, but managed to review data science material in R
- Revised notes from a previous Harvard Data Science in Genomics course that I had audited in Feb
- Spent time on reading papers in Computer Vision to further deepen my understanding of CNNs
- Read about CNN architectures in depth
- Ran the numbers and did some research for the Acko insurance hackathon proposal
- Spent time reviewing Deep Dream and Neural Style transfer with Gatys et al and their paper
- Worked on the documentation for the Acko Hack proposal of insurance premium predictions for self-driving car adoption
- Made plans for the home stretch of the challenge with lots of cool stuff planned for the 18 days to come and more!
- Worked on the neural style transfer assignment
- Read up more and worked on the implementation of the Neural Style Transfer paper from scratch
- Completed the course on CNNs as part of the Deep Learning Specialization Certificate
- Spent learning about the different clustering techniques apart from K-Means to solve the question for an internship interview
- Learnt about the end-to-end data science pipeline
- Spent a whole lot of time going through different papers handpicked from Arxiv and Arxiv sanity
- Topics and papers include CycleGan (Didn't really get it!), DeepFace, FaceNet..etc
- Started the Sequence Models course by Andrew NG
- Attended the Google Cloud Meetup which covered topics including using Kibana, elasticsearch and Cloud ML API
- Attended the GDG AI Meetup at Altimetrik
- Worked on the colab notebook pertaining to building a top class sentiment analyser using Spacy and Altair
- NLP with Spacy which is an industry grade NLP library along with NLTK
- Worked on time series analysis with Google cloud codelab
- GPU accelarated sessions with PyTorch support (CUDA backed)
- Worked on O'Reilly's Tensorflow from scratch challenge form the E-Mail newsletter
- Also had my DBMS lab exam today. Did great by the way!
- Worked on implementing RNNs and GRUs from scratch
- Understood the efficiency tradeoffs of GRUs while working on it
- Using the UCI FNA (Fine Needle Aspiartion) dataset to classify tumours
- Worked on implementing the entire Machine learning pipeline from data preprocessing to model validation
- Used pandas for data pre-processing, seaborn for exploratory data analysis and used a SVM
- Working on 3 Kaggle datasets to predict insurance claims in various categories
- Used the AllClaims Insurance dataset to precict the insurance claims in Auto and Life insurance industries
- Working on a chat application for better customer retention
- Working on the Acko insurance project submission
- Learnt about lstms and seq2seq word embeddings to build a chatbot
- All set for the hackathon tomorrow!
- Completed Week 1 of Andrew NG's Sequence Models course
- Working on the Jazz production with lstms problem statement
- Successfully submitted the Acko Hackathon solution (We didn't get shorlisted :( )
- Resumed work on my Genomics Research paper
- Learnt about Python galaxy for DNA Sequencing
- Spent time learning about Stanford's clusters for gene sequencing and their enormous budgets!
- Autoregressive Integrated Moving Average models are perfect for time series prediction
- Used it on data that includes a seasonal temporal shift. The data was non-stationary and had trends in the distribution and thus had to be integrated wth the differences as used in Box-Jenkins approach
- Walk-forward validation is extremely accurate as it provides every iteration with all the available data. This is computationally intensive and hence can used only for small datasets.
- Learnt about memory networks and applied it on the bAbI dataset
- Memory networks can also be used to make chatbots as they have more information gain than lstms with seq2seq embeddings
- Worked on the Reddit dataset to build a general purpose dataset
- Private repo for now. Will make it public soon!
- It has been a wonderful learning curve and am looking forward to do another one post my exams!