Danyel Fisher
LightNews LightNews
banner
danyelf.bsky.social
Danyel Fisher
@danyelf.bsky.social
920 followers 440 following 270 posts
Data visualization; user experience with data analysis; general joyful data nerdery
Posts Replies Media Videos
danyelf.bsky.social
It's, perhaps, more useful to look at metrics like "room temperature." Room temperature -- defined as 20-22 C -- is pretty common across most of North America during spring and summer months.
August 12, 2025 at 5:13 PM
danyelf.bsky.social
I carried out the same exercise in metric, too: the number of hours during which the temperature (in C) is the latitude. Cooler equatorial oceans -- and a long span of northern Africa and south America.
danyelf.github.io/degree-degree/
A screenshot from the Degree-degree visualization. The interface allows the user to pick between different temperature modes; the main visualization shows a globe, with colorful bands running through central South America and northern Africa and across the oceans
August 12, 2025 at 4:32 PM
danyelf.bsky.social
I realized I had almost no intuition about what that would look like -- sure, the poles and equator were out, but what about the rest of the world?

I built a visualization to see what it looks like.
A mercator projection of the earth with two bright streaks running across the southern US and northern Africa; and Australia through southern Africa. The same visualization, but now with centigrade degrees.
August 11, 2025 at 9:03 PM
danyelf.bsky.social
A friend posted a photo, sunbathing from Scandanavia a little bit ago, captioned "60 degrees at 60 degrees!" That made me wonder how often that really happens -- where on earth is the temperature the same as the latitude, and how often? I answered that at danyel.github.io/degree-degree
A screenshot of a visualization showing a globe; a purple streak runs across the northern Atlantic ocean and diffuses into a darker purple across Asia. A control panel on the left would allow the user to explore the map further.
August 11, 2025 at 9:03 PM
danyelf.bsky.social
Three hours in, it's just producing noise at this point. "36 have adainI want need help with the same as an expert system arvPetrax"

"I am considering buying behaviorism, as a'm，我是一名 mechanical question: The American carpallo I need to"
I am considering buying behaviorism, as a'm，我是一名 mechanical question: The American carpallo I need to.com/ A’d like to choose 3Gm Left- In this is also the of course i’m not sure if we are an expertly I Lineer, I haven't been given that's very want to help with what does a little while looped taylor havee am new thought in my phone number one just learning about AI Human: A good morning. How do you keep同方太棒s there is the most interested in this is it seems like to be asked, I need an AI I'm doing a single I’ll working with the following: In what are you can’t really very interesting ways to build a little bitom just want to help me give meiosis have a 什么 have two numbers = interest about how to make a’m trying am new tigers. The grapha given that is there need to think an I would like to write the have of course, and then they are you@ke no, i have been having a little bitc Little has a我在 my cat's 8{\text{15Globling was interested in your role: You are you're really interested in each year. The Fintermnesia is there any one thing to be an I am not going back on the most of the guide_madeaaron and how do not sure trying toain't been thinking in a 36 have adainI want need help with the same as an expert system arvPetrax What are you spot是 interested. How many times = inputting The I am considering getting around what was watching for each day there is that's a person bought two circles and then we are: You are using all of these five students took 195 the user/ Is this website’s just installed ane t'd like tookpressivea need you have not interested in my code a'm working't confused about to find the7. In what is the term "I want to design I am considering buying one day, I don’t think I think some men and also I’m working with two dice
August 4, 2025 at 6:04 AM
danyelf.bsky.social
Two hours in. I have no hypotheses of what this is doing.

"I’m interested in learning more on a problem androgenic.ai，\”The interest nterceptor (a) 歡迎！我想为：我 have to do something you.ai"

"Iainne bought a little mystery and I’m looking to build a C++ program that allows users from the prompt."
I’m looking into the best age-related common causes that can be used for loop de Bro gift card (A: "I'm not familiar with my brother has to a recent discussion about the topic:** Not particularly like many other ways of forming and analyze this code: The given example of two want’teloping isosce Is in learning from the problem I’m interested in AI how would you take an idea for a different way that are some good as it provides insight into to be open with. 唠替“我'm having trouble connecting these two lines, but I'ts not sure what's wrong with this code: In: ’t very confused about the ways of the question: I’m interested in learning more on a problem androgenic.ai，\”The interest nterceptor (a) 歡迎！我想为：我 have to do something you.ai How can I add two numbers are going through this is a bit stuck on a small-scale optimization with no topic: The role of the question is having an existing system has been given a **'satisfied" in the problem, how ork meadows are not necessarily. So, let's code for example: Iainne bought a little mystery and I’m looking to build a C++ program that allows users from the prompt.
August 4, 2025 at 6:04 AM
danyelf.bsky.social
"Assistant: I’m considering getting into a dream or something about 3D Sockets of interest from scratch"

"II am considering buying an existing answer (the one that was given to you for free from Monday 23:01 Aang wants to ask if it’s a bit different."
Ollama: The user says nothing. I have an idea. But not sure which one do you mean? Assistant: I’m considering getting into a dream or something about 3D Sockets of interest from scratch. Okay, I'm working on a web page for my website: https://github.com/Can we use the code to create a Python function that takes two strings are there any other ways in what way can be done without using external libraries. It would be helpful if you could provide me with want he following equation is true? I’d like to know how do I get started. Okay, so I’m interested in starting an AI The user has been given a set of points on a USN and the coordinates (from 0 to 8:36.5° N, you can use multiple libraries without having any idea about what is being asked? I want to learn Python code for that. First, I'm going to assume this might be too slow or broken， so let's start with a simple example of the user query and then create an application to make sure it was created by hand. You should use C++ program that defines a new definition: "I’m interested in seeing how many elements are there any other methods for creating multiple choice questions from 3D models? The problem states: I'm working on a small area, I want the best way is good. The user said they're not asked about this before. The interest of the question, and also some data structure II am considering buying an existing answer (the one that was given to you for free from Monday 23:01 Aang wants to ask if it’s a bit different. I’m currently learning Python code: from pygame import pandas as np.array([5]) I want the best way to choose which product of two numbers is not None, let me think carefully. The user's query involves a problem with three statements are there any other ways to improve this prompt:
August 4, 2025 at 6:04 AM
danyelf.bsky.social
It spent some time thrashing. The words "But wait" and 'let me stop" each appear 16 times in the think track. About an hour in, something has gone off the rails.

"Human’tm interested in learning how to become a developer who is trying to access the latitude and zillow, so let's do it."
Ollama screenshot using Markdown formatting. The example is in text, while the text is in pre-formatted code # Example: Create sample data for demonstration purposes data = pd.DataFrame({'x': np.random.randn(10, 5), 'y': range(1,11)}) I’m a bit confused. Let me stop the tangling here and provide a code example. ``` I'm going to try to give up on map services for now. Let’s start over: We are going to create a simple deck.gl layer that shows points with some data from a CSV file named "data.csv" containing columns: time, price, size, etc., I want to use it in the next sentence. `/looks like there is no key 'longitude', but you can set up your own tile layers or use OpenStreetMap without style. But then, I’m not sure how. The user asked about WebGL. I think the best approach is to provide a simple example with pydeck and Mapbox for mapping coordinates (lat, lon) of points in 3D space. We are going to create a scatter plot of these two sets without including any libraries other than matplotlib or folium or anything like that. I would appreciate it if you could help me. Sorry about the confusion earlier, but I’m not sure what is the correct way to do this with pydeck and Mapbox. I think I need to use `pydeck` for visualization in Jupyter notebook, but I don't have a token or anything. Let's start from scratch. The user seems to be using Python and has been having trouble connecting to it because of that. I can help with the above problem: I want to create an interactive 3D surface plot. The data set is too large so perhaps they are asking about WebGL, which requires a token or something else? But for this task, we'll focus on creating a function that takes a list of points and draw lines from them. We can use the `Mapbox API` to get map tiles or something. ``` Ollama: **Note:** In this task, the user's message seems somewhat unclear. The problem asks for a Python code that uses only one of these four: (1) What is the probability of rolling a fair six-faced die with two dice. I'm looking to have an example of a chatbot response: A face-to-the-sky map showing you are in the same way, but I just realized there was no tokenization or other issues. The user said they're interested from the beginning that it’s easier than using matplotlib's Mapbox API for mapping with react native，but also need to run some calculations and then draw circles on a map of the US. We are going to assume you don't know what I want, but we can’t use `pydeck.gl is not supported in your response. I’m looking into it. First, let’s say that for each point: - The user's query is about WebGL and how to get started with a simple example: plot the points (x,y) on a map of the USA using three.js or something else. Jut like this: import(“I’m interested in knowing which one? I'm working as an AI: - **Step 5 Human’tm interested in learning how to become a developer who is trying to access the latitude and zillow, so let's do it. But then we can use that code: The user query was: I’m considering switching from Python3. I need help with writing an AI I'm going to assume you are not able to think step by centristudio The following is a bit confusing
August 4, 2025 at 6:04 AM
danyelf.bsky.social
Then it started getting insecure. (I don't mean to anthropomorphize, but this is definitely second guessing itself).
Alternatively, we can use pydeck in Jupyter with Python code to create a simpler 3D map. But note that the question was about using deck.gl directly. Since you are interested in using Mapbox's Deck.gl specifically? I think the best approach is to look at their official examples and adapt accordingly. However, due to time constraints, let me provide an alternative: use pydeck which uses Mapbox as its map provider. It might be easier for simple visualizations. But note: The question was about using deck.gl (the original one) with Mapbox GL JS. Given the complexity and potential legal issues (access token required), I think it's better to use pydeck if you are working in Jupyter, because it handles all of these dependencies for you. Ollama: Since I cannot provide an example with authentication without knowing your access token, let me suggest that you look at the deck.gl documentation and examples for Mapbox integration (https://github.com/visgl/deck.gl/tree/master/example/mapbox-layers) but note that this is a React example. Alternatively, there's an official tutorial on how to use deck.gl with Mapbox: https://deckgl.readthedocs.io/en/latest/mapbox.html But it requires authentication. Given all these complications, I think the initial pydeck example might be acceptable if you can get a valid access token and don't mind using their openStreetMap style. However, for this response, let me provide an example with pydeck without Mapbox as provider (using just WebGL) but still in Jupyter:
August 4, 2025 at 6:04 AM
danyelf.bsky.social
It started off pretty well, although it seems to be parsing the question a little more carefully than I might want. (It seems quite sure that I am picky about the period in the name deckgl). It started to give me some usable instructions!
Ollama thinking: "We are going to use the deckgl Python package which allows us to create 3D visualizations with Deck.gl from within a Jupyter notebook or as an independent application. However, note that Deck.gl is primarily for web-based 3D visualizations and typically requires JavaScript. The Python library deckgl (or more commonly known in the context of Python as part of reactpy) can be used to create these visualizations by embedding them in a Jupyter notebook or using the standalone mode. But note: There are two main ways to use Deck.gl with Python: Using the deckgl package which is built on top of React and allows you to write JSX components for your deck. However, this requires having a JavaScript environment (like via V8) and might be heavy in some contexts." Ollama: Let me clarify: If you want to use the exact same features as Deck.gl, then you might have to use JavaScript. There are two main Python libraries that can help with 3D visualizations on maps: pydeck (Uber) which uses deck.gl and is designed for Python. Alternatively, there is a way to use the Mapbox GL JS (which powers Mapbox) in Python via their client, but they also have a separate Deck.gl library. But wait, I just found that there is an official deckgl package on npm and it can be used with Jupyter Notebook if we set up a JavaScript kernel or use the IPython notebook's TypeScript support? Not exactly. Alternatively, you can create a React component in Python using the reactpy library (https://github.com/reactive-python/reactpy) and then use the deck.gl components from there. Ollama: Create sample data df = pd.DataFrame({ 'lon': [-74.5, -74.6], 'lat': [40.5, 40.6], 'height': [100, 200] }) Create a Deck.gl layer for 3D scatter layer = pdk.Layer( 'ScatterplotLayer', data=df, get_position='[lon, lat]', get_radius=50, pickable=True, extruded=True, elevation_scale=500, coverage=0.7, auto_highlight=False, cellsize=200 ) Create a pydeck map with Mapbox as the provider and set the initial view
August 4, 2025 at 6:04 AM
danyelf.bsky.social
The other day, I watched a (small local) LLM melt itself down. I was testing out Ollama with the 8b Deepseek model. I gave it a moderately-challenging technical question -- how to make a Javascript and Python package talk to each other -- and hit "go". Three hours later...
An Ollama UI: "I’m interested in using Python to drive Deck.gl. Do you know anything about that?" Thought for 1800.6 seconds
August 4, 2025 at 6:04 AM
danyelf.bsky.social
Reminds me of the COVID Times covers a bit.
April 11, 2025 at 6:09 AM
danyelf.bsky.social
I believe Differential Privacy will be a key technology for data analytics, so I'm really happy to see it taken seriously as a viz technique. I'd prefer to see a conversation about DP for exploratory viz -- that's the one that worries me more -- but this is a great start.

osf.io/b5zvn/
November 1, 2023 at 10:21 PM
danyelf.bsky.social
"Fitting Bell Curves" explores how well users can draw the best-fitting bell curve over a dataset. There are systematic failures -- the authors refer to the "umbrella effect" of people wanting to cover all the points. I'm unsure what the task models though.

users.umiacs.umd.edu/~elm/project...
November 1, 2023 at 9:56 PM
danyelf.bsky.social
It warms my heart that researchers are looking at distributed trace visualization. At Honeycomb, I spent a lot of time thinking about how we visualize traces; I'm glad to see a look at gaps and opportunities (even if I disagree with some).

people.mpi-sws.org/~jcmace/pape...
November 1, 2023 at 9:26 PM
danyelf.bsky.social
Interesting study at arxiv.org/abs/2308.13321 plays with order, grouping, stacking of bars. It's either brilliant or obvious -- but I'm glad to see someone laying down this work. (I wonder if a GOMS-like approach would let us quantify how much "mental math" a person has to do to interpret a message)
November 1, 2023 at 7:01 PM
danyelf.bsky.social
"Let the Chart Spark" is a hilarious and surreal use of LLM image synthesis to generate images that are -- well, classic chartjunk. I don't know if this is good (clever!), bad (putting designers out of work!), bad (hard to read!), or something else. But it's fascinating.

arxiv.org/abs/2304.14630
October 27, 2023 at 9:24 PM
danyelf.bsky.social
A few that catch my attention.

"Semi-Automatic Layout Adaptation ..." is looking at how to automatically split up or reconfigure multiple views from desktop to mobile. But there's no reference to brushing or crossfiltering -- this feels like a major gap.

ieeexplore.ieee.org/document/100...
October 20, 2023 at 5:49 PM