Implementation of HTML repr for HighLevelGraph layers #7763

GenevieveBuckley · 2021-06-04T09:49:07Z

Here's an implementation of a repr_html() for the HighLevelGraph layer. It should give us a nice starting point to discuss what information we do or don't want displayed, and how we want it to look.

Closes HTML repr for HighLevelGraphs #7754
Tests added / passed
Passes black dask / flake8 dask / isort dask

GenevieveBuckley · 2021-06-04T09:53:35Z

Examples:

from dask.datasets import timeseries

ddf = timeseries().shuffle("id", shuffle="tasks").head(compute=False)
ddf.dask

import dask.array as da

x = da.ones((9,9), chunks=(3,3))
y = x.T
z = y[0:4,0:4]

z.dask

GenevieveBuckley · 2021-06-04T09:58:32Z

I've tried to keep the look similar to @jacobtomlinson's work here dask/distributed#4857 just for consistency (although it seems I've forgotten to add a little blue square next to each layer, oops).

TODO:

left align the second column in the tables
make the layer text the same grey colour as the highlevelgraph info text
check dependencies still look good even if there are a lot of them
add the little blue squares next to each layer

For discussion

I would like to shift all the layer specific stuff into a _repr_html_() on Layer (not just HighLevelGraph). But currently we're including the dependencies information in the layer info table... and only the high level graph knows what those dependencies are. Is there a clean way to resolve this? Ideally we don't want a difference between the layer info shown in the HTML repr from either the high level graph or a single layer.

jacobtomlinson

This is great, really exciting to see this.

I'm pleased to see the design resources being used, but I think we should try and make the HLG repr visually different from the cluster/client/scheduler/worker repr. If a user is scrolling through a long notebook looking for one or the other they should be able to identify them at a glance.

We also have some information about the width of each layer right? Perhaps we could show something visually to represent that. I'm thinking along the lines of the minimaps you see in many text editors these days.

GenevieveBuckley · 2021-06-04T10:24:57Z

I'm pleased to see the design resources being used, but I think we should try and make the HLG repr visually different from the cluster/client/scheduler/worker repr. If a user is scrolling through a long notebook looking for one or the other they should be able to identify them at a glance.

Huh, that's an interesting point to consider. Maybe let's turn those squares into circles & maybe use a different color?

We also have some information about the width of each layer right? Perhaps we could show something visually to represent that. I'm thinking along the lines of the minimaps you see in many text editors these days.

I don't know what you mean by width, could you clarify this? The layers are all pretty heterogeneous in terms of what attributes they have (but we can write special case logic or start to enforce more homogeneous descriptive attributes)

jacobtomlinson · 2021-06-04T10:31:47Z

This is the design reference I've been working from.

I'm keen to try and stay within the colours and shapes shown there in the various diagrams. I would be tempted to say let's specifically keep this diagram in mind when putting together this repr.

So as you say maybe something like little grey circles would work well? Maybe a blue circle for the HLG and grey for the layers?

jacobtomlinson · 2021-06-04T10:32:53Z

By width I was specifically referring to the n_parts attribute in DataframeIOLayer.

DataFrameIOLayer<name='make-timeseries-4a2ca177f118daf3d8a7db6ab747672c', n_parts=30, columns=None>

But I see what you mean, there isn't much consistency between the layers so this could be tricky.

jacobtomlinson · 2021-06-04T10:33:48Z

Also it looks like Layer has a get_dependencies method so that could be used for individual layer reprs.

dask/dask/highlevelgraph.py

Line 160 in 90fccab

def get_dependencies(self, key: Hashable, all_hlg_keys: Iterable) -> set:

jacobtomlinson · 2021-06-04T11:19:08Z

I threw together a quick mock up in Figma. This is kinda what I had in mind.

jacobtomlinson · 2021-06-04T11:22:07Z

That little graph icon can be included with

<svg width="76" height="71" viewBox="0 0 76 71" fill="none" xmlns="http://www.w3.org/2000/svg">
<circle cx="61.5" cy="36.5" r="13.5" fill="#F2F2F2" stroke="#1D1D1D" stroke-width="2"/>
<circle cx="14.5" cy="14.5" r="13.5" fill="#F2F2F2" stroke="#1D1D1D" stroke-width="2"/>
<circle cx="14.5" cy="56.5" r="13.5" fill="#F2F2F2" stroke="#1D1D1D" stroke-width="2"/>
<path d="M28 16L30.5 16C33.2614 16 35.5 18.2386 35.5 21L35.5 32.0001C35.5 34.7615 37.7386 37.0001 40.5 37.0001L43 37.0001" stroke="black" stroke-width="1.5"/>
<path d="M40.5 37L40.5 37.75L40.5 37.75L40.5 37ZM35.5 42L36.25 42L35.5 42ZM35.5 52L34.75 52L35.5 52ZM30.5 57L30.5 57.75L30.5 57ZM41.5001 36.25L40.5 36.25L40.5 37.75L41.5001 37.75L41.5001 36.25ZM34.75 42L34.75 52L36.25 52L36.25 42L34.75 42ZM30.5 56.25L28.0001 56.25L28.0001 57.75L30.5 57.75L30.5 56.25ZM34.75 52C34.75 54.3472 32.8472 56.25 30.5 56.25L30.5 57.75C33.6756 57.75 36.25 55.1756 36.25 52L34.75 52ZM40.5 36.25C37.3244 36.25 34.75 38.8243 34.75 42L36.25 42C36.25 39.6528 38.1528 37.75 40.5 37.75L40.5 36.25Z" fill="black"/>
<circle cx="28" cy="16" r="2.25" fill="#E5E5E5" stroke="black" stroke-width="1.5"/>
<circle cx="28" cy="57" r="2.25" fill="#E5E5E5" stroke="black" stroke-width="1.5"/>
<path d="M45.25 36.567C45.5833 36.7594 45.5833 37.2406 45.25 37.433L42.25 39.1651C41.9167 39.3575 41.5 39.117 41.5 38.7321V35.2679C41.5 34.883 41.9167 34.6425 42.25 34.8349L45.25 36.567Z" fill="#1D1D1D"/>
</svg>

mrocklin · 2021-06-04T13:41:35Z

This is a very fun conversation to lurk on :)

…

On Fri, Jun 4, 2021 at 6:22 AM Jacob Tomlinson ***@***.***> wrote: That little graph icon can be included with <svg width="76" height="71" viewBox="0 0 76 71" fill="none" xmlns="http://www.w3.org/2000/svg"> <circle cx="61.5" cy="36.5" r="13.5" fill="#F2F2F2" stroke="#1D1D1D" stroke-width="2"/> <circle cx="14.5" cy="14.5" r="13.5" fill="#F2F2F2" stroke="#1D1D1D" stroke-width="2"/> <circle cx="14.5" cy="56.5" r="13.5" fill="#F2F2F2" stroke="#1D1D1D" stroke-width="2"/> <path d="M28 16L30.5 16C33.2614 16 35.5 18.2386 35.5 21L35.5 32.0001C35.5 34.7615 37.7386 37.0001 40.5 37.0001L43 37.0001" stroke="black" stroke-width="1.5"/> <path d="M40.5 37L40.5 37.75L40.5 37.75L40.5 37ZM35.5 42L36.25 42L35.5 42ZM35.5 52L34.75 52L35.5 52ZM30.5 57L30.5 57.75L30.5 57ZM41.5001 36.25L40.5 36.25L40.5 37.75L41.5001 37.75L41.5001 36.25ZM34.75 42L34.75 52L36.25 52L36.25 42L34.75 42ZM30.5 56.25L28.0001 56.25L28.0001 57.75L30.5 57.75L30.5 56.25ZM34.75 52C34.75 54.3472 32.8472 56.25 30.5 56.25L30.5 57.75C33.6756 57.75 36.25 55.1756 36.25 52L34.75 52ZM40.5 36.25C37.3244 36.25 34.75 38.8243 34.75 42L36.25 42C36.25 39.6528 38.1528 37.75 40.5 37.75L40.5 36.25Z" fill="black"/> <circle cx="28" cy="16" r="2.25" fill="#E5E5E5" stroke="black" stroke-width="1.5"/> <circle cx="28" cy="57" r="2.25" fill="#E5E5E5" stroke="black" stroke-width="1.5"/> <path d="M45.25 36.567C45.5833 36.7594 45.5833 37.2406 45.25 37.433L42.25 39.1651C41.9167 39.3575 41.5 39.117 41.5 38.7321V35.2679C41.5 34.883 41.9167 34.6425 42.25 34.8349L45.25 36.567Z" fill="#1D1D1D"/> </svg> [image: Group 2] <https://user-images.githubusercontent.com/1610850/120793973-790d9480-c52f-11eb-88a5-06611efdce48.png> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#7763 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACKZTHHERQDKJJQHYDIM3TTRCZO3ANCNFSM46CO647Q> .

GenevieveBuckley · 2021-06-07T02:51:02Z

What I really want is to have the is_materialized info for all layers available at a glance. Ideally I could make the icon be either a solid or an empty circle, depending on status.
Hm, I had been assuming the get_dependencies method would materialize the graph. In retrospect, I don't have a good reason for assuming that, other than the fact lots of innocuous seeming things seem to cause unwanted materialization. I'll check!
Your icon with the html snippet is pretty cute! I'll try adding it in and we'll see how it looks.

GenevieveBuckley · 2021-06-07T04:28:12Z

Update: it turns out that result.dask.dependencies[layer_key] is not exactly equivalent to dask_layer.get_dependencies(). The get_dependencies method appears to return more low level information.

Here's a short example using the timeseries dataset:

from dask.datasets import timeseries
ddf = timeseries().shuffle("id", shuffle="tasks").head(compute=False)
layer_key = 'simple-shuffle-029255f702ea9cf087d6955375c8e9ad'

This is what we get when we look at the high level graph dependencies

>>> ddf.dask.dependencies[layer_key]
{'make-timeseries-2d58a4d4763bdeb3eca63e8b5ffd6c3b'}

But when we use the layer get_dependencies method, there are two things that are different

We can't use the same string as the key anymore, we need a tuple key instead; and
The results are more low-level information than we saw previously

>>> layer = ddf.dask.layers[layer_key]
>>> tuple_key = ('simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0)
>>> layer.get_dependencies(tuple_key, layer.keys())

{('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 0),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 1),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 2),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 3),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 4),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 5),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 6),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 7),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 8),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 9),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 10),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 11),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 12),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 13),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 14),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 15),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 16),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 17),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 18),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 19),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 20),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 21),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 22),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 23),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 24),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 25),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 26),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 27),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 28),
 ('split-simple-shuffle-029255f702ea9cf087d6955375c8e9ad', 0, 29)}

GenevieveBuckley · 2021-06-07T08:40:58Z

Starting to look nice:

The default is to have the "Layers" bit expanded, so all the layer headings are visible but not the detailed layer information.

TODO:

make the layer text the same grey colour as the highlevelgraph info text - DONE. They match now, although I've chosen a darker grey that passes a contrast accessibility check (join the discussion here)
~~add the little blue squares next to each layer~~ Add grey icons like Jacob suggests - DONE
make the layer icons look different depending on whether they are materialized or not (light grey = unmaterialized; dark grey = materialized) - DONE
~~left align the second column in the tables~~ It turns out that this is being overridden by jupyter's CSS settings (at least, that's why stack overflow says my align tag is being ignored), so I think we should probably leave this alone.
check dependencies still look good even if there are a lot of them - working on this now. It doesn't really look great imo, there's a lot of mostly-redundant information

jacobtomlinson · 2021-06-07T08:56:12Z

dask/highlevelgraph.py

+        0px;">{highlevelgraph_info}</p>
+
+        <div style="">
+            <details open style="margin-left: 0px;">


I did not know you could default them to open, that's awesome!

... I didn't know that either until I really wanted to do it! I'm very happy that turned out to be easy

Sadly, this "details open" trick is causing problems with the test checking HTML validity. So I think we'll just get rid of it entirely for now.

GenevieveBuckley · 2021-06-07T09:03:05Z

What I'm thinking about now is how to get the layer names from within the layers themselves. From the high level graph, I just used the dictionary key for that particular layer. I'm not totally sure what the canonical way to get that information from inside the layer is yet.

There's no requirement for layers to have a .name attribute, and so many of them don't have this info stored nicely. For example, anything that is a MaterializedLayer doesn't have a name.

mrocklin · 2021-06-07T14:45:56Z

For tests I recommend that we ...

Verify that the HTML is clean (I think that there are dashboard plots

mrocklin · 2021-06-07T14:48:29Z

Whoops, errant send, trying again

For tests I recommend that we ...

Verify that the HTML is clean, see distributed.utils.is_valid_xml
Check that certain terms occur in the output, like the names of the types of a layer or two.

Clearly we can't test the visual aesthetics, but having a couple of basic tests here is likely to keep this code healthy long term

mrocklin · 2021-06-07T14:51:15Z

I think that it would be good to have a nicer string here

I wonder if there is a nicer way to show this information.

For example, saying is_materialized: True seems clinical. It's how we would write it out in code. Instead, we might have an icon somewhere for this. We might also include the number of nodes in each layer.

These things could be in a table, they could also be in the summary. We can stuff more information there for at-a-glance information.

dask/highlevelgraph.py

mrocklin · 2021-06-07T15:50:34Z

dask/highlevelgraph.py

+        "is_materialized": layer.is_materialized(),
+        "dependencies": dependencies,
+    }
+    return info


Not important, but my preference would be to keep logic like this inlined in order to avoid indirection (blog post here: https://matthewrocklin.com/blog/work/2019/06/23/avoid-indirection )

However, really I'd prefer that we move this down to Layer subclasses (even more indirection, I acknowledge).

I agree, it should go with the Layer.

I don't agree that it should be inlined with the _repr_html_ function logic, because imo that will make it harder for people making representations for layer subclasses. (Perhaps I misunderstand your point here though)

For someone writing a _repr_html function for a Layer subclass, it seems a lot easier to grab the default dictionary of interesting layer information with this function,, then add more dictionary keys/values specific to that particular Layer subclass, and then turn that into a nice HTML string. If you only have the HTML string, you might end up either duplicating a lot of the logic here for each layer subclass, OR faffing around removing HTML div tags, and trying to insert the new extra information into the right place in the HTML string.

I agree that dupliation would ensure (we would have to address this with methods, functions, etc..

However it could open up larger variations than just different keys/values in the table. For example we might include images like the chunk structure, or different coloring based on layer type.

GenevieveBuckley · 2021-06-08T02:20:39Z

We might also include the number of nodes in each layer.

Maybe this is a silly question, but how do I know what the number of nodes in a layer is?

jsignell · 2021-06-10T14:13:04Z

I love this! I feel like it's already helping me gain insight. I pushed it a little using different chunking with Matt's example:

import dask.array as da

x = da.random.random((10000, 10000), chunks=(100, 100))
y = x + x.T - x.mean(axis=0)

Some ideas:

Having the same summary items as arrray makes a lot of sense to me - this is the same as Matt's 1 above, but maybe including shape in there as well?
The label area of the repr table seems wider than it needs to be.
The size of the headers feels large to me. Especially when compared to the dask.array repr.

dask/highlevelgraph.py

Co-authored-by: Julia Signell <[email protected]>

…ey/dask into HighLevelGraphs-repr-html

GenevieveBuckley · 2021-06-11T01:46:19Z

Here's what it looks like now

Dataframe example

Array example

Note: there's something odd on my machine making the little circle icons not line up with the headings. I also don't seem to get the little arrows indicating you can expand the collapsible details. However, Jacob, Matt, and Julia all have really pretty screenshots in this thread, so I think it's probably fine for everyone else.

GenevieveBuckley · 2021-06-11T01:51:46Z

Sorry to keep coming back with suggestions

Don't be sorry, it's fun to tinker.

In some more aggressive future we might consider including the dask array SVG array chart directly in the HTML, that's probably future work though

Sure, sounds good.

Here's a summary of the changes I've made in the last round of edits:

fixed bug showing dtype as None where it shouldn't
add array shape to collection_annotations
add array chunksize to collection annotations
Add columns to dataframe collection annotations
Add npartitions to the dataframe collection annotations
Removed verbose array chunks information from collection_annotations (we can always add this back in if we need it later)
Removed verbose dataframe divisions information from collection_annotations (we can always add this back in if we need it later)

Here's what didn't change:

Matt suggests: For class types can we try repr(cls) rather than str(cls) so that we get something like the following:
```
In [4]: da.Array
Out[4]: dask.array.core.Array
```
Answer: I'm sorry, I can't make this render without the <class '...'> brackets (which is what I think you're asking me to do). I tried these combinations of things in collection_annotations, and all of them render the same way in the output table:
```
"type": self.__class__,
"type1": str(self.__class__),
"type2": repr(self.__class__),
"type4": type(self),
"type5": str(type(self)),
"type6": repr(type(self)),
```
Julia suggests: making the font size for the headings smaller.
Sure, I'm happy to implement this is there is agreement on what size it should be. I used some of the choices from Jacob's HTML work here Add HTML repr to scheduler_info and incorporate into client and cluster reprs distributed#4857 as a guide, so we'd want to be consistent everywhere, if possible.

mrocklin · 2021-06-11T02:28:43Z

Answer: I'm sorry, I can't make this render without the <class '...'> brackets (which is what I think you're asking me to do). I tried these combinations of things in collection_annotations, and all of them render the same way in the output table:

Yeah, I just tried this myself. I expected it to work because of this:

I suspect that the IPython display function must have special cased this, causing the confusion.

mrocklin · 2021-06-11T02:30:47Z

I'm happy to withdraw my request.

mrocklin · 2021-06-11T02:31:14Z

Tests seem to be sad. Happy to merge otherwise though

…phs-repr-html

jrbourbeau · 2021-06-11T22:10:51Z

Just merged main to make tests happy

jrbourbeau · 2021-06-11T22:14:57Z

For class types can we try repr(cls) rather than str(cls) so that we get something like the following

FWIW we have a typename utility function which should help with this:

In [1]: import dask.array as da

In [2]: from dask.utils import typename

In [3]: typename(da.Array)
Out[3]: 'dask.array.core.Array'

jacobtomlinson · 2021-06-14T09:24:26Z

I see what @jsignell means about the titles. Perhaps for the layers we should drop the size down to <h4>?

jsignell · 2021-06-14T14:21:17Z

I think h4 for the layers would be nice, but I am also totally happy with this going in as-is. It is a vast improvement over

mrocklin · 2021-06-14T22:52:58Z

Let's merge this in. We can iterate.

GenevieveBuckley · 2021-06-15T02:46:20Z

I didn't check what it looks like when the jupyter notebook is in dark mode, that's probably another useful thing to try.

jacobtomlinson · 2021-06-15T09:19:26Z

Most of the text is readable. The darker gray isn't so good and the icons need updating.

jacobtomlinson · 2021-06-15T09:44:30Z

I've opened #7809 to add dark mode support.

Builds on #7763 and adds dark mode support.

_repr_html_() for HighLevelGraph layers

e83bb6a

jacobtomlinson reviewed Jun 4, 2021

View reviewed changes

GenevieveBuckley mentioned this pull request Jun 7, 2021

Add convenience method is_materialized to check all high level graph layers #7762

Closed

2 tasks

Better icons for HTML repr of HighLevelGraph

bd16878

jacobtomlinson reviewed Jun 7, 2021

View reviewed changes

mrocklin reviewed Jun 7, 2021

View reviewed changes

dask/highlevelgraph.py Outdated Show resolved Hide resolved

mrocklin reviewed Jun 7, 2021

View reviewed changes

GenevieveBuckley added 5 commits June 8, 2021 13:39

Add test for valid HTML

d48c6fd

Fix xml import

a68b14d

Move html repr onto Layer, add test for HTML validity

d4bceab

Fix comments

026eaaa

Split tests so the minimum dependency CI environments don't fail

89cb3c7

github-actions bot added the array label Jun 8, 2021

jsignell reviewed Jun 10, 2021

View reviewed changes

dask/highlevelgraph.py Outdated Show resolved Hide resolved

GenevieveBuckley and others added 3 commits June 11, 2021 09:12

Update dask/highlevelgraph.py

e87ae42

Co-authored-by: Julia Signell <[email protected]>

Update HTML repr table for layers

121d8ec

Merge branch 'HighLevelGraphs-repr-html' of github.com:GenevieveBuckl…

5168052

…ey/dask into HighLevelGraphs-repr-html

GenevieveBuckley and others added 3 commits June 11, 2021 15:33

Fix failing argtopk test (meta input was incorrect)

bfc289a

Simpler fix for the tests

0a3ce66

Merge branch 'main' of https://github.com/dask/dask into HighLevelGra…

db9806b

…phs-repr-html

jrbourbeau mentioned this pull request Jun 14, 2021

Release 2021.06.1 dask/community#165

Closed

4 tasks

Merge branch 'main' into HighLevelGraphs-repr-html

efea19e

mrocklin merged commit ca94cb9 into dask:main Jun 14, 2021

jacobtomlinson mentioned this pull request Jun 15, 2021

Add dark mode support to HLG HTML repr #7809

Merged

mrocklin pushed a commit that referenced this pull request Jun 15, 2021

Add dark mode support to HLG HTML repr (#7809)

19df781

Builds on #7763 and adds dark mode support.

GenevieveBuckley mentioned this pull request Jun 16, 2021

Small tweaks to the HTML repr for Layer & HighLevelGraph #7812

Merged

3 tasks

GenevieveBuckley mentioned this pull request Jul 1, 2021

Adding new characteristics to the HLG visualizations #7847

Closed

2 tasks

GenevieveBuckley mentioned this pull request Jul 13, 2021

Add dask.array SVG to the HTML Repr #7886

Merged

2 tasks

dhomeier mentioned this pull request Aug 23, 2021

Extend __str__ and __repr__ methods of BaseTableCoordinate, ExtraCoords, GlobalCoords sunpy/ndcube#453

Merged

Implementation of HTML repr for HighLevelGraph layers #7763

Implementation of HTML repr for HighLevelGraph layers #7763

Conversation

GenevieveBuckley commented Jun 4, 2021 • edited Loading

GenevieveBuckley commented Jun 4, 2021

GenevieveBuckley commented Jun 4, 2021

TODO:

For discussion

jacobtomlinson left a comment

Choose a reason for hiding this comment

GenevieveBuckley commented Jun 4, 2021

jacobtomlinson commented Jun 4, 2021 • edited Loading

jacobtomlinson commented Jun 4, 2021

jacobtomlinson commented Jun 4, 2021

jacobtomlinson commented Jun 4, 2021

jacobtomlinson commented Jun 4, 2021

mrocklin commented Jun 4, 2021 via email

GenevieveBuckley commented Jun 7, 2021

GenevieveBuckley commented Jun 7, 2021

GenevieveBuckley commented Jun 7, 2021

jacobtomlinson Jun 7, 2021

Choose a reason for hiding this comment

GenevieveBuckley Jun 7, 2021

Choose a reason for hiding this comment

GenevieveBuckley Jun 8, 2021

Choose a reason for hiding this comment

GenevieveBuckley commented Jun 7, 2021

mrocklin commented Jun 7, 2021

mrocklin commented Jun 7, 2021

mrocklin commented Jun 7, 2021

mrocklin Jun 7, 2021

Choose a reason for hiding this comment

GenevieveBuckley Jun 8, 2021

Choose a reason for hiding this comment

mrocklin Jun 8, 2021

Choose a reason for hiding this comment

GenevieveBuckley commented Jun 8, 2021

jsignell commented Jun 10, 2021

GenevieveBuckley commented Jun 11, 2021 • edited Loading

Dataframe example

Array example

GenevieveBuckley commented Jun 11, 2021

mrocklin commented Jun 11, 2021

mrocklin commented Jun 11, 2021

mrocklin commented Jun 11, 2021

jrbourbeau commented Jun 11, 2021

jrbourbeau commented Jun 11, 2021

jacobtomlinson commented Jun 14, 2021

jsignell commented Jun 14, 2021

mrocklin commented Jun 14, 2021

GenevieveBuckley commented Jun 15, 2021

jacobtomlinson commented Jun 15, 2021

jacobtomlinson commented Jun 15, 2021

GenevieveBuckley commented Jun 4, 2021 •

edited

Loading

jacobtomlinson commented Jun 4, 2021 •

edited

Loading

GenevieveBuckley commented Jun 11, 2021 •

edited

Loading