From 4e0b2350f2052f6d3bdd07187415d32485b4e76a Mon Sep 17 00:00:00 2001
From: Nan Xiao <me@nanx.me>
Date: Sat, 26 Oct 2024 23:30:43 -0400
Subject: [PATCH] Remove less relevant text for blog post

---
 docs/articles/get-started.md  | 32 +-------------------------------
 docs/articles/get-started.qmd | 31 +------------------------------
 2 files changed, 2 insertions(+), 61 deletions(-)

diff --git a/docs/articles/get-started.md b/docs/articles/get-started.md
index 5839181..e9ef1aa 100644
--- a/docs/articles/get-started.md
+++ b/docs/articles/get-started.md
@@ -11,31 +11,7 @@
     python3 examples/get-started.py
     ```
 
-## Introduction
-
-Fitting topic models at scale using classical algorithms on CPUs can be
-slow. Carbonetto et al. (2022) demonstrated the equivalence between
-Poisson NMF and multinomial topic model likelihoods, and proposed a
-novel optimization strategy: fit a Poisson NMF via coordinate descent,
-then recover the corresponding topic model through a simple
-transformation. This method was implemented in their R package,
-[fastTopics](https://cran.r-project.org/package=fastTopics).
-
-Building on this theoretical insight, tinytopics adopts an alternative
-approach by directly solving a sum-to-one constrained neural Poisson
-NMF, optimized using stochastic gradient methods, implemented in
-PyTorch. Although this approach may not have the same theoretical
-guarantees, it does have a few potential practical benefits:
-
-- Scalable: Runs efficiently on both CPUs and GPUs and enables
-  large-scale topic modeling tasks.
-- Extensible: The model architecture is flexible and can be extended,
-  for example, by adding regularization or integrating with other data
-  modalities.
-- Minimal: The core implementation is kept simple and readable,
-  reflecting the package name: **tiny**topics.
-
-This article shows a canonical tinytopics workflow using a simulated
+Let’s walk through a canonical tinytopics workflow using a synthetic
 dataset.
 
 ## Import tinytopics
@@ -170,9 +146,3 @@ plot_top_terms(
 ```
 
 ![](images/get-started/F-top-terms-learned.png)
-
-## References
-
-Carbonetto, P., Sarkar, A., Wang, Z., & Stephens, M. (2021).
-Non-negative matrix factorization algorithms greatly improve topic model
-fits. arXiv Preprint arXiv:2105.13440.
diff --git a/docs/articles/get-started.qmd b/docs/articles/get-started.qmd
index 3d160c8..807bf2a 100644
--- a/docs/articles/get-started.qmd
+++ b/docs/articles/get-started.qmd
@@ -14,30 +14,7 @@ eval: false
     python3 examples/get-started.py
     ```
 
-## Introduction
-
-Fitting topic models at scale using classical algorithms on CPUs can be slow.
-Carbonetto et al. (2022) demonstrated the equivalence between Poisson NMF
-and multinomial topic model likelihoods, and proposed a novel optimization
-strategy: fit a Poisson NMF via coordinate descent, then recover the
-corresponding topic model through a simple transformation.
-This method was implemented in their R package,
-[fastTopics](https://cran.r-project.org/package=fastTopics).
-
-Building on this theoretical insight, tinytopics adopts an alternative
-approach by directly solving a sum-to-one constrained neural Poisson NMF,
-optimized using stochastic gradient methods, implemented in PyTorch.
-Although this approach may not have the same theoretical guarantees,
-it does have a few potential practical benefits:
-
-- Scalable: Runs efficiently on both CPUs and GPUs and enables large-scale
-  topic modeling tasks.
-- Extensible: The model architecture is flexible and can be extended,
-  for example, by adding regularization or integrating with other data modalities.
-- Minimal: The core implementation is kept simple and readable, reflecting
-  the package name: **tiny**topics.
-
-This article shows a canonical tinytopics workflow using a simulated dataset.
+Let's walk through a canonical tinytopics workflow using a synthetic dataset.
 
 ## Import tinytopics
 
@@ -168,9 +145,3 @@ plot_top_terms(
 ```
 
 ![](images/get-started/F-top-terms-learned.png)
-
-## References
-
-Carbonetto, P., Sarkar, A., Wang, Z., & Stephens, M. (2021).
-Non-negative matrix factorization algorithms greatly improve topic model fits.
-arXiv Preprint arXiv:2105.13440.