Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
Tags
- clinical-ml 8
- fastapi 3
- python 3
- devops 3
- aws 3
- xgboost 3
- gradient-boosting 3
- aws-ecs 2
- ecs-fargate 2
- docker 2
- alb 2
- networking 2
- shap 2
- explainability 2
- engineering 1
- oauth 1
- debugging 1
- systems 1
- mcp 1
- backend 1
- ai 1
- mlops 1
- explainable-ai 1
- gam 1
- pharma 1
- mlflow 1
- github-pages 1
- dns 1
- namecheap 1
- blogging 1
- infrastructure 1
- microservices 1
- codepipeline 1
- codebuild 1
- ecr 1
- bedrock 1
- nextjs 1
- flask 1
- secrets-manager 1
- gbdt 1
- hyperparameters 1
- regularization 1
- early-stopping 1
- subsample 1
- lightgbm 1
- goss 1
- efb 1
- leaf-wise 1
- catboost 1
- categorical-features 1
- ordered-boosting 1
- target-encoding 1
- feature-importance 1
- waterfall 1
- beeswarm 1
- class-imbalance 1
- smote 1
- adasyn 1
- borderline-smote 1
- cost-sensitive 1
- imbalanced-learn 1
- sklearn 1
- pipeline 1
- column-transformer 1
- feature-engineering 1
- custom-transformer 1
- synthetic-data 1
- ctgan 1
- tvae 1
- sdv 1
- privacy 1
- healthcare-ml 1
- data-generation 1
clinical-ml
Feature Engineering Pipelines: Building Preprocessing That Can’t Leak
Scaling your test data with statistics from test data is a bug. So is fitting your imputer on training + test combined. sklearn Pipelines prevent both - here...
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
CatBoost: How It Handles Categorical Features Without Leaking the Answer
The standard way to encode categorical features (target encoding) secretly leaks information from the labels into the training data. CatBoost fixes this with...
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
XGBoost Hyperparameters Explained: What Each Knob Actually Does
n_estimators, max_depth, subsample, lambda, gamma - not as a grid to search, but as levers with specific, observable effects. Run each experiment and watch t...
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
fastapi
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
python
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
devops
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
aws
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
xgboost
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
XGBoost Hyperparameters Explained: What Each Knob Actually Does
n_estimators, max_depth, subsample, lambda, gamma - not as a grid to search, but as levers with specific, observable effects. Run each experiment and watch t...
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
gradient-boosting
CatBoost: How It Handles Categorical Features Without Leaking the Answer
The standard way to encode categorical features (target encoding) secretly leaks information from the labels into the training data. CatBoost fixes this with...
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
aws-ecs
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
ecs-fargate
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
docker
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
alb
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
networking
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
shap
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
explainability
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
engineering
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
oauth
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
debugging
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
systems
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
mcp
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
backend
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
ai
Reverse Engineering an MCP-Based Job Application (Without Instructions)
How I cracked an unconventional job application that gave me nothing but a URL
mlops
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
explainable-ai
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
gam
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
pharma
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
mlflow
How I Built an Explainable AI System to Optimize Pharma Sales Calls
When a client said ‘no black boxes’, I discovered GAM - and it changed how I think about enterprise ML
github-pages
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
dns
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
namecheap
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
blogging
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
infrastructure
From GitHub Pages to a Real Domain - How I Finally Claimed edwardpraveen.com
A step-by-step account of moving my technical blog off a github.io subdomain, the architecture decisions I weighed, and exactly what I clicked (and skipped) ...
microservices
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
codepipeline
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
codebuild
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
ecr
Deploying 5 Microservices to AWS ECS - A Complete Hands-On Journey from GitHub to Production
A complete, hands-on walkthrough of deploying five FastAPI microservices and a React frontend to AWS ECS Fargate using CodeBuild and CodePipeline, with path-...
bedrock
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
nextjs
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
flask
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
secrets-manager
Deploying Two AI Applications on AWS ECS Fargate: A Journey from Local Dev to Production
A detailed, hands-on walkthrough of deploying two LLM-powered applications (EvalAgent Studio and AI Test Driven) on AWS ECS Fargate, detailing architectural ...
gbdt
Understanding Gradient Boosting Internals: From Pseudo-Residuals to XGBoost
Most people use XGBoost without understanding what it’s actually computing. This post walks through the internals - pseudo-residuals, learning rate, regulari...
hyperparameters
XGBoost Hyperparameters Explained: What Each Knob Actually Does
n_estimators, max_depth, subsample, lambda, gamma - not as a grid to search, but as levers with specific, observable effects. Run each experiment and watch t...
regularization
XGBoost Hyperparameters Explained: What Each Knob Actually Does
n_estimators, max_depth, subsample, lambda, gamma - not as a grid to search, but as levers with specific, observable effects. Run each experiment and watch t...
early-stopping
XGBoost Hyperparameters Explained: What Each Knob Actually Does
n_estimators, max_depth, subsample, lambda, gamma - not as a grid to search, but as levers with specific, observable effects. Run each experiment and watch t...
subsample
XGBoost Hyperparameters Explained: What Each Knob Actually Does
n_estimators, max_depth, subsample, lambda, gamma - not as a grid to search, but as levers with specific, observable effects. Run each experiment and watch t...
lightgbm
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
goss
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
efb
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
leaf-wise
LightGBM: Why It’s Faster Than XGBoost and When That Matters
LightGBM trains 10–20x faster than XGBoost on large datasets. This post explains the three algorithmic tricks behind that speed - GOSS, EFB, and leaf-wise gr...
catboost
CatBoost: How It Handles Categorical Features Without Leaking the Answer
The standard way to encode categorical features (target encoding) secretly leaks information from the labels into the training data. CatBoost fixes this with...
categorical-features
CatBoost: How It Handles Categorical Features Without Leaking the Answer
The standard way to encode categorical features (target encoding) secretly leaks information from the labels into the training data. CatBoost fixes this with...
ordered-boosting
CatBoost: How It Handles Categorical Features Without Leaking the Answer
The standard way to encode categorical features (target encoding) secretly leaks information from the labels into the training data. CatBoost fixes this with...
target-encoding
CatBoost: How It Handles Categorical Features Without Leaking the Answer
The standard way to encode categorical features (target encoding) secretly leaks information from the labels into the training data. CatBoost fixes this with...
feature-importance
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
waterfall
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
beeswarm
SHAP From Scratch: From Feature Importance to Per-Patient Explanations
feature_importances_ tells you which features the model used most globally. SHAP tells you why the model made a specific prediction for a specific patient. H...
class-imbalance
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
smote
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
adasyn
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
borderline-smote
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
cost-sensitive
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
imbalanced-learn
Class Imbalance in Clinical ML: SMOTE, ADASYN, and Why Accuracy Is Lying to You
A model that predicts ‘no readmission’ for every patient can claim 88% accuracy. Here’s how SMOTE, ADASYN, BorderlineSMOTE, and cost-sensitive learning actua...
sklearn
Feature Engineering Pipelines: Building Preprocessing That Can’t Leak
Scaling your test data with statistics from test data is a bug. So is fitting your imputer on training + test combined. sklearn Pipelines prevent both - here...
pipeline
Feature Engineering Pipelines: Building Preprocessing That Can’t Leak
Scaling your test data with statistics from test data is a bug. So is fitting your imputer on training + test combined. sklearn Pipelines prevent both - here...
column-transformer
Feature Engineering Pipelines: Building Preprocessing That Can’t Leak
Scaling your test data with statistics from test data is a bug. So is fitting your imputer on training + test combined. sklearn Pipelines prevent both - here...
feature-engineering
Feature Engineering Pipelines: Building Preprocessing That Can’t Leak
Scaling your test data with statistics from test data is a bug. So is fitting your imputer on training + test combined. sklearn Pipelines prevent both - here...
custom-transformer
Feature Engineering Pipelines: Building Preprocessing That Can’t Leak
Scaling your test data with statistics from test data is a bug. So is fitting your imputer on training + test combined. sklearn Pipelines prevent both - here...
synthetic-data
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
ctgan
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
tvae
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
sdv
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
privacy
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
healthcare-ml
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...
data-generation
Synthetic Data Generation for Healthcare: CTGAN, TVAE, and When to Use Each
When you can’t share real patient data for model development, synthetic data is the answer - if it’s done right. This post covers CTGAN, TVAE, and how to eva...