Machine Learning
AWS Machine Learning Specialty
AWS| May 2022
Demonstrated ability to build, train, tune, and deploy ML models using the AWS Cloud.
Can derive insight from AWS ML services using either pre-trained models or custom models built from open-source frameworks.
Open Source frameworks include Spark, Zepplin, and Hadoop.
Machine Learning DevOps Engineer Nano Degree
Udacity | Feb 2022
This Nano degree focuses on writing production-level models.
Consists of 5 different modules, each module has a specific project to test your abilities.
Clean Code Principles
Building a Reproducible Model Workflow
Deploying a Scalable ML Pipeline in Production
ML model scoring & Monitoring
Tools Learnt:
Git & Github
Weights and Biases
Ml flow
Hydra
Pytest
Conda
Yaml
Docker
DVC
AWS
Heroku
FAST API
Flask
WSL
Crontab
Skills Developed:
- Clean Code Principles
Writing results.log, using logger for debugging
Writing README.md files for Project reusability
Model development in git
Modularisation of Project Libraries & Function
Using Pylint to score your code out of 10
Using Autopep8 to fix issues regarding formating of your code
Skills Developed:
2. Building a reproducible workflow
Creating a clean, organized, reproducible, end-to-end machine learning pipeline from scratch using conda, Hydra and Pytest.
Transforming, validating and verifying your data
Writing Deterministic tests and non-deterministic tests for various datasets.
Use fixtures to share data between tests.
Use conftest.py to add options to the command line of pytest so you can pass parameters and use it within components of ML pipelines.
Running experiments, tracking data, code, and results using tags.
Understanding of Feature Engineering and Feature store.
Creating an inference pipeline with sklearn and export it with mlflow and scikit-learn and Pytorch.
Evaluate the inference artifact against the test dataset
Selecting the best performing model and promoting it to production.
Creating an artifact ready for deployment, and releasing your final pipeline on Github.
Assign version numbers using Semantic Versioning
Deploy inference artifact with MLflow and other tools for both online (real-time) and offline (batch) inference
Skills Developed:
3. Deploying a Scalable Machine Learning Model
Review Validation Sets and K-Fold Cross-Validation.
Data Slicing
Data Slicing Use Cases and Testing
Model Bias
The Aequitas Package
Model Cards
Data Provenance
Remote Storage with DVC
Pipelines with DVC
Experiment Tracking with DVC
Continuous Integration with GitHub Actions
Continuous Deployment with Heroku
Fundamentals of FastAPI
Local API Testing
Live API Testing
Skills Developed:
4. ML Model Scoring & Monitoring
Writing Model files to persistent storage and production directories on Amazon S3 database
Automation of Job Scheduling with crontab
Scoring models using relevant analysis functions
Reading and writing model scores
Model drift, reepelacing obsolete models and hypothesis testing
Latency consideration in model re-deployment pipeline
Data Integrity testing
Data Instability testing
Updating outdated dependencies automatically
API configuration with Python flask
End-point scripting
Calling API's
GET & POST
Tensorflow Developer Certificate
Tensorflow Developer Program- Google Developer Certification| Jan 2022
Earned by giving a 5 hour exam of modelling neural networks in Tensorflow framework in python. The certificate program requires an understanding of building TensorFlow models using Computer Vision, Convolutional Neural Networks, Natural Language Processing, Sequences and time series, and real-world image data and strategies.
Skills Developed:
Foundational principles of ML and Deep Learning
Building ML models in TensorFlow 2.x
Building image recognition, object detection, text recognition algorithms with deep neural networks and convolutional neural networks
Using real-world images in different shapes and sizes to visualize the journey of an image through convolutions to understand how a computer “sees” information, plot loss and accuracy
Exploring strategies to prevent overfitting, including augmentation and dropouts
Applying neural networks to solve natural language processing problems using TensorFlow
Tools Learnt:
Tensorflow
Keras
Python
Pandas
Numpy
Deep Learning Specialisatiton
Deep Learning.AI| Dec 2020
This Certificate consists of completing 5 different courses in Deep learning topic
Courses include:
Neural Networks & Deep Learning
Improving Deep Neural Networks: Hyperparameter Tuning, Regularisation and Optimisation
Convolutional Neural Network
Structuring Machine Learning Projects
Sequence Models
Each courses required project assigment to be completed described in the project section.
Tools Learnt:
Python
Jupyter Notebooks
Pandas
Numpy
Tensorflow
Skills Developed:
- Neural Networks & Deep Learning
Binary Classification
Logistic Regression
Cost Functions
Gradient Descent
Vectorization
Using Numpy
Activation Functions
Sigmoid
Relu
Leaky Relu
Tanh
Softmax
Forward Propagation
Backward Propagation
2. Improving Deep Neural Networks: Hyperparameter Tuning, Regularisation and Optimisation
Bias and Variance in Deep Learning
Data Segregation
Train/Valid/Test sets splits
Regularisation
Dropout
Data Augmentation
Early stopping
Normalizing Inputs
Vanishing/ Exploding Gradients
Weight Initialization of Neural Networks
Gradient Checking
Batch Normalization
Multiclass Classification
Softmax activation function
Optimization Algorithms
Gradient Descent
Stochastic Gradient Descent(SGD)
Batch Gradient Descent
Gradient Descent withMomentum
Mini-batch gradient descent
RMS Prop
Adam
Learning rate decay
Hyperparameters Tuning
Momentum value (Optimizer)
Beta (Optimizer)
Layers
Hidden units
Learning rate decay
Mini-batch size
Skills Developed:
3. Convolutions Neural Network
Convolution Operation
Components in convolutional Neural network
Filters
Strides
padding
Pooling Layers
Average Pooling
Max Pooling
ResNets
Inception Networks
Transfer Learning
Data Augmentation
Object Localisation
Bounding box prediction
Non max supression
Intersection over union
Anchor Boxes
YOLO algorithm
R-CNN
Fast R-CNN
Faster R-CNN
Face Recognition
One shot learning
Siamese Network
Triplet Loss function
Face Verification
Neural Style Transfer
Content cost function
Style cost function
4. Structuring Machine Learning Projects
Avoidable bias
Single number evaluation metrics
Data segregation on large datasets
Error Analysis
Mismatched training and dev/test set
Human Level Performance
Transfer Learning
Multi Task Learning
Skills Developed:
5. Sequence Models
Recurrent Neural Network
Back Propagation through time
Different types of RNN
many-to-many
many-to-one
one-to-one
one-to-many
Vanishing Gradient in RNN
Gated Recurrent Unit(GRU) Layers
Long short term memory (LSTM) Layers
Bi-directional RNN
Deep RNN
Word Embeddings
Learning Word Embeddings
Word2Vec
Negative Sampling
GloVe
Application using Word Embeddings
Sentiment Classification
Debiasing Word Embeddings
Various Sequence to Sequence Architecture
Beam Search
Bleu Score
Attention Model
Machine Learning
Stanford Online| Oct 2020
This course teaches about the concepts applied in machine learning.
Tools Learnt
MATLAB
Skills Developed:
Programming Fundamentals
Supervised Learning
Linear Regression
Logistic Regression
Unsupervised Learning
K-means algorithm
Elbow method
Neural Networks
Forward Propagation
Backward Propagation
Support Vector Machines
Model Performance Evaluation
Regularisation
Principal Components Analysis
Anomaly Detection
Recommender Systems
Photo OCR - Object Detection