Synthetic Musings

Unfiltered technical write-ups, model implementation notes, and practical machine learning research and random thoughts. Part lab notebook, part technical archive.

Date Title Description Reading Time
Sep 17, 2026 Who Does the Clearing? Two mechanisms that give the same mean clearance curve, a history-dependent recurrent-event model that separates them, and the design that makes the contrast causal rather than merely predictive. 24 min
Sep 15, 2026 CP or Tucker CP returns components that are, under Kruskal’s condition, the only ones that fit. Tucker returns a basis per mode and a small core, fixed only up to a rotation you can slide between them. The question you are asking decides which answer is useful. 9 min
Sep 13, 2026 Probabilistic PCA Derive the PPCA covariance fit and latent posterior, then compare its shrunk reconstruction with classical PCA on a reproducible sensor example. 9 min
Sep 13, 2026 Six Views of PCA Derive the links between variance, projection error, covariance eigenvectors, and truncated SVD. MovieLens and metabolomics show why the largest-variance direction still needs interpretation. 31 min
Sep 10, 2026 GAPDH as a NET suppressor Li et al. 2023 show that flow through GAPDH suppresses neutrophil extracellular traps through intracellular pH, not extra NADPH. The public ST002477 table cannot carry that claim. Companion to Neutrophil Metabolomes as a Matrix. 21 min
Sep 10, 2026 Neutrophil Metabolomes as a Matrix Seventy-five neutrophil metabolomes, three COVID labels. k-means, Ward, and a diagonal GMM with k=3 do not recover the labels. Averaging the groups and fitting a line through those means is well-defined and the wrong summary. 18 min
Sep 10, 2026 A NumPy-like Tensor in Pure Python A flat list plus a shape tuple is enough for shape, ndim, size, len, reshape, transpose, and indexing — no NumPy. Reshape keeps storage order; transpose does not. 19 min
Sep 4, 2026 What the Immune System Sees in an AAV Vector A recombinant AAV particle carries no gene the immune system can read as therapeutic. Four arms answer it — innate sensing of the shell and the DNA, antibodies that block or tag it, T cells that kill the cells that took it up, and complement — and a route widget here lets you switch each one on and watch the surviving genomes fall. 17 min
Sep 3, 2026 Why So Many Matrix Factorizations Different factorizations exist because they solve different problems: QR for tall least squares, Cholesky for an SPD covariance, LU for a square system that is not SPD, SVD for a rank-k summary or a rectangular inverse. 16 min
Aug 31, 2026 Claude Agent SDK in Python query() and ClaudeSDKClient wrap Claude Code so you do not write the Messages API tool loop by hand. 11 min
Aug 31, 2026 Fingerprint Algorithms: From Ridge Formation to On-Device Embeddings Friction ridges are a prenatal pattern. Identification is the problem of turning that pattern into a comparable digital object — first as minutiae, now as a learned embedding bounded by the sensor. 46 min
Aug 31, 2026 Tensor Inverses: Independence, Completion, Identification You already undo a mixed recording, fill a blank rating, and pick plant settings from a desired output. For a square, full-rank matrix those three jobs share one inverse. A… 19 min
Aug 30, 2026 Tensor Inverses in Practice Four real problems where the operator is a tensor. For each one, this post names the product that applies, takes the inverse in that product, and measures what flattening… 34 min
Aug 29, 2026 Claude API with the Anthropic Python SDK The official anthropic Python SDK talks to Claude through one method, client.messages.create(). Roles, content blocks, tools, and retrieved documents are fields on that call… 12 min
Aug 29, 2026 Claude Certified Architect, Week 1: The Orchestrator-Worker Loop Week one of preparing for the Claude Certified Architect – Foundations exam: the four-stage agent lifecycle, a production orchestrator-worker loop that branches on every stop_reason, and the failure modes that separate a demo from a service. 20 min
Aug 29, 2026 GitHub vs GitLab: Origins, Dominance, and AI-Native Hosts GitHub and GitLab differ by who built them, who owns them, and who leads today — and four AI-native platforms are now reshaping git hosting around them. The comparison… 3 min
Aug 29, 2026 Uses of Tensor Factorizations The main application of tensor factorizations is compression: images and video, or the parameter tensors inside a neural net — a VGG conv kernel, a transformer dense map. 50 min
Aug 28, 2026 NumPy to JAX: The Array That Learned New Tricks NumPy stores numbers in arrays whose operations run in compiled code. JAX keeps that array API and adds compilation, automatic differentiation, and accelerator backends. 3 min
Aug 28, 2026 Tensor Factorizations and Tensor Inverses A tensor \(\mathcal{X}\in\mathbb{R}^{I_1\times\cdots\times I_N}\) is an \(N\)-way array. The \(N=2\) case is a matrix: rows and columns only. 15 min
Aug 27, 2026 From Dataset to Biological Signature Golub’s 1999 leukemia data, run end to end: recovering the original train/test split from a CSV that lost its labels, fitting a gene signature without letting the test data leak in, and showing why the test-set error rate is the hard one to get. 16 min
Aug 27, 2026 What is a Biological Signature? A fingerprint identifies a person from a patch of skin. A biological signature does it with molecules — and pulling one out of data takes four steps, of which the last is the one people get wrong. 2 min
Aug 22, 2026 The Cheapest Ladder Is Also the Shortest Maths costs a pencil and tells you when you’re right, which is why it’s full of broke kids. Then the bill arrives at twenty-two — and the way out asks for something a pencil never needed: a quiet room. 5 min
Aug 19, 2026 The Matrix That Rotates, Stretches, and Rotates Again Every matrix turns space, stretches it, and turns it again — and that one fact prices image compression in decibels per byte, predicts when a solver will return garbage, hides inside PCA and topic models, and pulls a voice back out of equal-parts noise. With two widgets you can drive yourself. 34 min
Aug 19, 2026 The Anatomy of a Volcano Plot Why statistical significance does not imply biological importance, how the log2 transform makes the two directions comparable, and how a pair of thresholds turns thousands of noisy hypotheses into a shortlist you can defend. With a screening simulation you drive yourself. 13 min
Aug 18, 2026 Five Photos of One Kite, Five Different Shapes Kendall shape analysis quotients out translation, scale, and rotation (the similarity group). Field photographs of a rigid object introduce perspective, which similarity… 18 min
Aug 18, 2026 The Statistical Jackknife Quenouille’s bias eraser and Tukey’s pseudovalues, worked through four examples: ratio bias in small cohorts, promo-abuse syndicates, censored survival regression, and clinical biomarker stability — with interactive widgets that run in the browser. 14 min
Aug 14, 2026 Everyone’s Name on Every Paper Paper and citation counts credit every name on the author list. They do not measure work done. 5 min
Aug 14, 2026 What’s Left When You Take Away Position, Size and Angle Shape is what remains after removing translation, scale, and rotation from a landmark configuration. The removal steps are deterministic (no fitting). The resulting shape… 15 min
Aug 13, 2026 The Directions a Matrix Refuses to Turn Multiplying a vector by a matrix \(A\) generally rotates and scales it. 5 min
Aug 12, 2026 Explainability is Not Causality: What is not Why Attribution methods localize: they say which part of a fixed model carried a fixed input to a fixed output. Causal claims are bought by intervening. The two fields share a vocabulary — why, because, counterfactual, important — and mean different things by all four. 8 min
Aug 12, 2026 Matrix Factorizations as Optimization Problems: QR, SVD, Eigendecomposition, NMF, and Cholesky A matrix factorization writes \(A\) as a product of simpler matrices (triangular, orthogonal, diagonal, or all-nonnegative). Each standard factorization is also the solution… 10 min
Aug 10, 2026 How Causality Works: From Toddlers to Do-Calculus Perception, covariation, intervention, counterfactuals — children assemble a causal toolkit before school. Pearl’s graphs and Rubin’s potential outcomes are that toolkit written down, and on a growth team’s promo and feature they turn two dashboard numbers into two different decisions. 15 min
Aug 10, 2026 Wiring an Eval Pipeline to the Claude API “Reply with JSON only” is a request, not a guarantee. Structured outputs turn the rubric into a contract the API enforces — and once the verdict is a typed object, the rest of the pipeline is a batch job and two ratios. 8 min
Aug 10, 2026 Metagenomics: Reading the Life That Won’t Grow For a century microbiology studied whatever would grow on a petri dish — a fraction of a percent of what was in the sample. Metagenomics threw away the dish: extract all the DNA at once, sequence it, and let the organisms stay uncultured. 4 min
Aug 9, 2026 How Prompt Evaluation Pipelines Work A demo is a sample of size one, and the qualities you care about — tone, helpfulness, groundedness — have no regex. That leaves another model holding the grading pen, which works only if you measure how often it agrees with you. 6 min
Aug 9, 2026 Cascade or Native? What the Voice AI Stack Actually Looks Like in 2026 Native speech-to-speech models beat cascaded pipelines on latency by seconds. What open native models have not closed is the reasoning gap. The cascade still sells a transcript you can read before the user hears anything, at a price the benchmarks make uncomfortably clear. 19 min
Aug 8, 2026 LLM API Pricing Compared: Claude, GPT, Gemini, and the Chinese Challengers (August 2026) Per-token rates across seven providers as of August 2026 — and the four structural differences (caching, long-context surcharges, modality coverage, output multiples) that move real bills further than the headline number does. 10 min
Aug 8, 2026 Waiting, and Not Waiting The Messages API is one endpoint with two delivery modes — and the difference stops being cosmetic the moment tool calls enter the picture. 5 min
Aug 6, 2026 Stop Parsing Prose Asking an LLM for JSON in English often returns JSON wrapped in extra text: a preamble, markdown fences, or a trailing remark. json.loads() rejects all of those. 4 min
Aug 5, 2026 The One Thing Always in the Room If perception is prediction all the way up, the self may be the invariant the hierarchy eventually notices in itself — and the loop that produces it is lossy, so it cannot be run backwards. 7 min
Aug 5, 2026 Can You Invert a Recursive Function? A function is invertible if and only if it is injective: distinct inputs produce distinct outputs. 3 min
Aug 3, 2026 Built for Problems It Has Never Seen Skill at a known task can be bought with data. Sample-efficient generalization to a novel one cannot. That conversion is the human niche, and tools are the physical record of it. 2 min
Aug 3, 2026 Your LLM Has No Memory. Everything Else Is Engineering. An LLM is stateless per call. What we call memory is a set of choices about what goes into the prompt, what gets stored outside it, and when it comes back. 7 min
Aug 3, 2026 From Formants to Foundation Models Every era of TTS moved the unit of generation down a level of abstraction — from rules, to recorded fragments, to parameters, to raw samples, to tokens. You can hear each step. 13 min
Aug 2, 2026 LLM Agents from First Principles An LLM agent is a while loop with one stochastic step per iteration. Everything else — message history, tool dispatch, output parsing, turn budgets — is deterministic Python… 6 min
Jul 31, 2026 Explainability Is a Localization Problem Four attribution methods on a misclassified MNIST digit and four classical decompositions of Fisher’s irises, implemented from scratch and checked at build time — set against one framing that covers all eight. 14 min
Jul 30, 2026 Nvidia Bought the Team Behind the Library I Contribute To Nvidia has bought Kumo AI, which employs most of the people who maintain PyTorch Geometric. This is what that acquisition looks like from inside the library. 5 min
Jul 28, 2026 The Professor Who Let Us Break R First Brian Ripley has just shared the 2026 Rousseeuw Prize for building and maintaining R. This is a note on what that work looked like from inside his statistics laboratory… 5 min
Jul 28, 2026 Swarms in Worktrees: How Claude Code Agents Work in Parallel Parallel coding agents sharing one checkout overwrite each other’s edits. Git worktrees give agents separate working directories over shared history, eliminating file-level… 5 min
Jul 27, 2026 ‘It Works on My Machine’: The Ladder of Numerical Dependencies Underneath numerical code sits a ladder of dependencies — architecture, driver, runtime, packages — and pinning one rung leaves every other rung free to move. 4 min
Jul 26, 2026 Why the Bayesian Bootstrap? Rubin’s Dirichlet weights derived from scratch, set against Efron’s multinomial counts, on S&P 500 and IBM returns and on MNIST — with every closed form checked against simulation at build time. 76 min
Jul 24, 2026 The Poor Person’s Bayesian Rubin’s Dirichlet weights, Efron’s multinomial counts, fifteen law schools, and the exact point at which resampling stops being inference on the cheap. 73 min
Jul 24, 2026 Skills Ate Commands: What Changed in Claude Code Claude Code stores user prompts as markdown under .claude/commands/ (slash commands) or .claude/skills/ (skills). Custom commands merged into skills; the commands reference n… 13 min
Jul 24, 2026 Lost in Translation: How Math Tribes Speak Different Languages (And Why Epidemiologists Hate MSE) The word calibration names different procedures in statistics, epidemiology, machine learning, and mathematics. Cross-disciplinary work fails when the same term is used… 12 min
Jul 23, 2026 From First Principles to Interior Point Methods: The Physics of Constrained Convex Optimization Convex optimization minimizes a convex objective over a convex feasible set. For such problems: 16 min
Jul 23, 2026 LangGraph vs LlamaIndex: One Agent, Two Architectures Multi-step LLM workflows need control flow beyond a single chat() call. LangGraph models agents as graph traversal over shared state; LlamaIndex Workflows models them as even… 13 min
Jul 23, 2026 Training vs. Calibrating Epidemiological Models: A First-Principles Guide Training: minimize prediction error (e.g. MSE) for dynamical parameters. 46 min
Jul 23, 2026 Maximum-Margin Classification from First Principles: SVMs as Convex Optimization A support vector machine (SVM) selects a separating hyperplane by solving a convex optimization problem that maximizes geometric margin. Soft-margin SVMs add slack penalties… 21 min
Jul 23, 2026 What TDA Actually Builds: A Filtered Simplicial Complex, Not a Fancy Graph TDA builds a filtered simplicial complex over a fixed vertex set — not a threshold graph alone. Graphs cannot distinguish a hollow cycle from a filled triangle; simplicial… 19 min
Jul 23, 2026 Uses and Abuses of Topological Data Analysis: A Pragmatic Guide Density methods (k-means, KDE, GMM) answer where mass concentrates. Topological data analysis (TDA) answers connectivity: gaps, loops, tunnels, voids. A filled disk and an… 15 min
Jul 22, 2026 Two Ways to Expand the Space: Persistent Homology and Kernel SVMs on the Same Toy Dataset A disk inside a ring is not linearly separable in \(\mathbb{R}^2\). Persistent homology (unsupervised) reports the annular hole. A kernel SVM (supervised) finds a curved… 6 min
Jul 22, 2026 Topological Data Analysis: Finding Clusters, Loops, and Voids Across Scales Clustering output depends on a scale hyperparameter (k, eps, dendrogram cut height). Topological data analysis (TDA) indexes every scale at once via a filtration and records… 9 min
Jan 8, 2026 Thoughts on Unsupervised Learning The unsupervised methods arranged as one family tree, with compression as the trait they share, and clustering as its extreme case. 3 min
Apr 9, 2025 AI Engineer vs ML Engineer vs Data Scientist: Roles, Tools, and Code Quality An AI Engineer, an ML Engineer and a Data Scientist do three different jobs, and the titles blur enough that a job advert, a teammate, or a piece of code can be hard to place. 6 min
Apr 9, 2025 Cursor Is VS Code With the Model Inside the Editor What I found when I installed Cursor to see what the fuss was about: the editor I already used, with completion, inline editing and chat that read the codebase. As of April 2025. 3 min
Apr 9, 2025 Model Evaluation Metrics Two classifiers with identical accuracy, one missing twice as many cancers. A metric is a decision about which errors you will tolerate. 8 min
Apr 8, 2025 Classifying PDF Documents with RAG and LLMs You can classify chunks or classify documents, but not both. The aggregation step you skip past is where the actual modelling decision lives. 11 min
Apr 7, 2025 Apple Silicon Broke an Assumption the ML Toolchain Was Built On Why Docker, PyTorch and conda each rub on an ARM Mac, and the native route that avoids most of it. As of spring 2025. 5 min
Apr 6, 2025 Hugging Face Hub Python SDK Every Hub repo is a git repo, which means every unpinned from_pretrained is a dependency on a moving target. 6 min
Apr 6, 2025 Accelerating Transformers with Hugging Face Optimum Exporting to ONNX is a compile step: you trade the ability to change the model for a graph a dedicated runtime can optimise. Here is what that trade actually pays. 7 min
Apr 6, 2025 Transformers Library for NLP pipeline() is three objects behind one call — and fine-tuning is what forces you to take them apart. 9 min
Apr 5, 2025 Introduction to Diffusers by Hugging Face A diffusion pipeline is four separable parts. Every way of controlling the output — prompts, inpainting, ControlNet — is an intervention on a different one. 10 min
Apr 4, 2025 Hugging Face Datasets Why load_dataset returns instantly on a dataset larger than your RAM, and what that buys you downstream. 8 min
Apr 4, 2025 Efficient Few-Shot Learning with SetFit from Hugging Face SetFit trains a classifier from 16 labelled examples in under a minute. On one GLUE task that beats the baseline by 21 points; on another it loses to always guessing. 8 min
Apr 2, 2025 Snowflake Bills Storage and Compute Apart, and That Is the Whole Design How separating storage from compute changes what a warehouse costs and how it scales, with the Python and SQL to drive it. As of spring 2025. 4 min
Apr 1, 2025 Prompting Is Still Authorship, and the Checking Is the Cost An honest account of how I have used AI tools to draft posts on this blog: what the trade bought, what it cost, and what changed when the corpus was reread. 3 min
Mar 31, 2025 Airflow Earns Its Keep at the Third Retry When a scheduled script stops being enough and a workflow orchestrator starts paying for itself: the handful of Airflow concepts that matter, and what each replaces. As of early 2025. 4 min
Mar 31, 2025 Models Rot Silently, and Drift Is Not the Same Failure as Bias Why a deployed model gets worse without changing, the two different ways it fails, and which of AWS’s two SageMaker tools watches which. As of early 2025. 5 min
Mar 30, 2025 Soft Prompts vs. Hard Prompts Initialise a soft prompt from real words and it sits exactly on the vocabulary. Sixty gradient steps later it is somewhere no sentence can reach. 7 min
Mar 27, 2025 Exploring Vector Databases: Weaviate vs Pinecone Weaviate and Pinecone are two vector databases. They store and query vector embeddings, numerical representations of text, images, or audio. That is the index behind semantic… 4 min
Mar 27, 2025 PySpark Is for Data That Does Not Fit on One Machine What Spark does that pandas cannot, how it keeps a distributed computation fast and recoverable, and the small example that shows each mechanism. As of early 2025. 4 min
Mar 26, 2025 LangGraph Turns an LLM Pipeline Into a Graph With One Shared State What a graph of nodes over one mutable state buys over a linear chain: branching, parallel work, and a workflow that remembers where it is. The 2025 tour; the API has moved since. 4 min
Mar 22, 2025 dbt Turns a Pile of SQL Scripts Into a Project What changes when SQL transformations get references, tests, documentation and version control: the dbt setup end to end on Amazon Redshift. As of early 2025. 5 min
Mar 21, 2025 An Embedding Makes Similarity a Distance What text, image and voice embeddings have in common, how each modality learns to place similar things near each other, and what the geometry does and does not promise. 5 min
Mar 20, 2025 Hugging Face Tokenizers Library A tokenizer’s vocabulary is learned, not configured — and when it is learned from too little text, it fails in a way you can read off the output. 6 min
Mar 16, 2025 LangChain: Introduction LangChain 1.x deleted LLMChain and ConversationBufferMemory. What replaced them is not a new API — it is the observation that neither needed to be a class. 7 min
Mar 16, 2025 Parameter-Efficient Fine-Tuning with Hugging Face’s peft Library LoRA trains 0.24% of GPT-2’s parameters and saves to a 1 MB file — the reason that works is a claim about the update, not about the model. 6 min
Mar 12, 2025 Evaluating NLP Models with the Hugging Face evaluate Library The same prediction scores 0.0 and 0.96 depending on which metric you load. evaluate makes that swap one line — which is the danger, not the feature. 8 min
Mar 11, 2025 Databricks Is Managed Spark, and the Extras Are the Reason to Pay What Databricks adds to a Spark cluster you could run yourself: notebooks on a shared cluster, Delta Lake’s transactions, and MLflow’s run tracking. Working notes, as of early 2025. 4 min
Mar 3, 2025 A Docker Image Is a Stack of Diffs, and the Order Decides the Build Time Why a machine-learning image rebuilds in seconds or in twenty minutes, and a model served from a container end to end. 6 min
Mar 2, 2025 Kafka Moves Events, NiFi Routes Them, Flink Computes on Them Three streaming tools that are constantly mistaken for one another do three different jobs: a durable log, a visual router with provenance, and a stateful processor. Merges three earlier posts. 6 min
Feb 25, 2025 Data Types: Structured, Unstructured, and Semi-Structured Data Different data types, their characteristics, file formats, and applications in modern data processing 8 min
Feb 21, 2025 Quarto Workshop An overview of Quarto, its features, and how it can be used for reproducible scientific writing, blogging, and more. 3 min
Feb 20, 2025 Feature Importance After Clustering Once clustering is complete, understanding which features differentiate each cluster is crucial for interpretation. This blog post explores multiple ways to attribute… 2 min
Feb 9, 2025 First Job in Industry Getting hired in the first job 1 min
Feb 7, 2025 Python Books I Recommend, and How to Read One So It Sticks The Python books I would hand to someone at each level, with what I thought of each, and the one habit that makes a programming book work. 3 min
Feb 6, 2025 Reinforcement Learning Splits on One Question: Learn the Value, or Learn the Policy? A return to reinforcement learning for someone who left it for a while: the pieces of the problem, the two families of method and why they exist, where the field stands beside generative models, and the papers that built it. 9 min
Feb 4, 2025 A Method’s Name Is a Contract What the leading underscore, the decorator and the double underscores each promise to the next person who reads a Python class, and what to call instance, class, static, helper and dunder methods. 4 min
Feb 3, 2025 Retrieval-Augmented Generation Is Nearest-Neighbour Search Wired Into the Prompt What RAG adds to a language model, why the retrieval step bounds the answer, and how a vector store finds the neighbours fast enough to matter. Merges three earlier posts. 8 min
Feb 2, 2025 Conda Exists Because Scientific Python Is Not Python Why pip alone cannot install the numerical stack reliably, what conda does about it, what it costs, and where uv and pixi have moved the line since. As of early 2025. 4 min
Jan 31, 2025 Attention Is a Softmax Over Dot Products, and the Divisor Keeps It Trainable How a transformer decides which tokens matter to each other: the three matrices, the softmax that does the choosing, an interactive look at why the scores are divided by the square root of the width, and the layer in PyTorch. 6 min
Jan 27, 2025 Clustering Has No Labels to Tune Against, So the Metric Can Lie Which clustering hyperparameters can honestly be tuned on a held-out split, what the label-free metrics measure, and a worked case where the silhouette picks the wrong number of clusters with a straight face. 7 min
Jan 21, 2025 Markdown: What’s in the Name? Why it is called markdown, and the design rule hiding in the name. 1 min
Sep 21, 2024 R Fits a Model When You Create It; scikit-learn Waits to Be Asked The same linear regression through both APIs, to the same score, and what the difference in when fitting happens rests on: S3 dispatch on one side, a class with fit and predict on the other. Either style can be imitated in the other language. 10 min
Sep 20, 2024 The Test Set Is Spent Once Why a dataset is cut into training, development and test sets, what leaks the test set before you ever use it, and how to make the split in the common libraries. Merges two earlier posts. 9 min
Sep 19, 2024 Model Selection Is Natural Selection, and the Validation Set Is the Habitat One analogy pushed as far as it goes: models as species, data as the environment, tuning as speciation, and the validation score as fitness. Where it holds, where it breaks, and the one thing it predicts. 6 min
Apr 22, 2024 grep Searches the Working Tree, git log Searches Time A real bug hunt through PyTorch Geometric’s repository, run in April 2024: two strings from a maintainer’s note, and the five commands that find where each one lives, lived, or never was. 6 min
Apr 11, 2024 A Sparse Tensor Stores the Nonzeros, and the Saving Has to Be Counted, Not Read Off COO and CSR on a mesh graph from ModelNet10, in PyTorch: what each format stores, why the obvious memory measurement reports no saving at all, and what a 700-fold matmul speedup did and did not compare. Run in April 2024. 10 min
Apr 9, 2024 Make Is for the Commands You Type, CI Is for the Ones You Forget A Makefile and a GitHub Actions workflow split a project’s chores by who has to remember them, with this blog’s own pipeline as the worked example. 5 min
Mar 3, 2024 PyTorch Geometric Installs on Apple Silicon, and Its Test Suite Does Not A March 2024 record of getting a developer install of PyTorch Geometric onto an M1 Pro: the wheels that just work, the C++ extensions that have to be built from source, and the test run that segfaults on collection. 7 min
Feb 9, 2024 Post With Code and Plot Let’s see if blogging with code and plots works here: 1 min
Feb 6, 2024 Welcome To Synthetic Musings This is the first entry in Synthetic Musings, where I write up what I am working through in machine learning, statistics, and the tooling around them. I’m Ravi (@project-delp… 1 min
No matching items