anndata_dask_array.ipynb: copy tweaks / improvements

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png

[ryan-williams]

anndata_dask_array.ipynb

• Seed da.random/np.random ⟹ allow regenerating the notebook deterministically.
• Copy tweaks: some typos, some more opinionated rephrasing, couple places described things that didn't match the cell outputs (e.g. claiming a slice result was eager when it wasn't), couple places I added a cell and some copy.
• Kernel name: python3.bak → python3

Dockerfile / regenerate.sh

Regenerate notebook files deterministically (in Docker / using juq to clean notebooks / canonicalize outputs), e.g.:

./regenerate.sh anndata_dask_array

I've used it here on anndata_dask_array.ipynb, but not any other notebooks.

[review-notebook-app]

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[flying-sheep]

Hi, thank you very much, this is a great idea. We really need to standardize a reproducibility story here, thank you for getting started here. I don‘t think pip install will install the same versions without pip-timemachine or a lockfile, so this is not completely reproducible. It would be amazing though if we found a way to get pixel-perfect graphics that don‘t generate a diff every time.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

[flying-sheep]

Hia, I don’t know if you saw my comments above: Could you please take a look?

[ryan-williams]

Sorry, I finally got back to this.

One thing about your solution: The font rendering is kind of ugly. Can you check if installing ipycytoscape instead of graphviz (and maybe also if setting dask’s config visualization.engine to it) improves this?

ipycytoscape embeds "widgets" in the notebook, which render as placeholder text when I reload the notebook:

I'm assuming they also won't render when compiled into a docsite either. lmk if I missed something there.

On the Graphviz side, Dask's to_graphviz hard-codes fontname="helvetica", but it renders differently for me on macOS vs. Ubuntu (and slightly differently still in an Ubuntu Docker image on an Ubuntu host, which I used in this PR). The existing images seem to match what I see when I run it on my Macbook, and both Ubuntu versions have "uglier" fonts. Here are samples:

macOS

Ubuntu Docker (this PR)

Ubuntu

Let me know how you want to proceed. I was also thinking I should add an example of reading from Dask (based on Scanpy's Dask tutorial, that @ivirshup pointed me at)

"One-liner" I used for extracting images (for reference, incl. my own)

r=copy # Git ref
f=anndata_dask_array.ipynb # notebook path
git show $r:$f \
| jq -r '.cells[] | (.outputs // [])[].data.["image/png"] | select(.)' \
| head -n1 \
| base64 -d \
>$r.png