Qgrnn updates #207

albi3ro · 2021-01-25T23:23:02Z

Some minor updates to the QGRNN tutorial. The updates include

separating weights and bias into two separate variables
use of numpy.rng
use of nx.complete_graph
Move optimization printing out of cost function
Use step_and_cost instead of just step
renaming some variables for readability

More complete work could transform the tutorial to using ApproxTimeEvolution, instead of a custom implementation of it.

josh146 · 2021-01-26T04:10:18Z

Thanks @albi3ro! Do the changes allow the demo to run in tape-mode?

albi3ro · 2021-01-26T04:18:04Z

It ran in tape mode with no changes and still does.

But executing it got me reading it. And reading it gave me ideas of things to change.

josh146 · 2021-01-26T04:26:27Z

Awesome 🙂 I assumed it was a required change for the demo to build with the new PL version. Since it is not required for supporting the new PL version, no need to prioritize review until after release -- I'll remove the requests for review I made.

glassnotes · 2021-01-26T14:56:45Z

@albi3ro out of curiosity did you record execution times for tape mode vs. non-tape mode? This is one of the demos that's currently non-executable because of its long runtime, however I noticed while testing some of the other demos that using tape mode resulted in some speedup.

albi3ro · 2021-01-26T15:18:18Z

Ok! That was significantly faster in tape mode. Basically an order of magnitude faster.

I ran for ten steps and timed using /usr/bin/time from the command line.

Non-tape mode was:
real 242.65
user 243.89
sys 30.96

Tape mode:
real 19.06
user 22.03
sys 11.96

glassnotes · 2021-01-26T15:25:59Z

Ok! That was significantly faster in tape mode. Basically an order of magnitude faster.

Incredible!! Thanks for looking into that. When we tackle #201 then we may be able to make this one executable 🎉

glassnotes · 2021-02-22T17:25:00Z

@albi3ro if you have the bandwidth, did you want to take a crack at making this one executable? (I am also happy to do so)

glassnotes

@albi3ro changes look great, what an amazing speedup! I just left a few minor comments, but otherwise this looks good to go.

glassnotes · 2021-02-24T23:01:38Z

demonstrations/tutorial_qgrnn.py

+    :property="og:description": Using a quantum graph recurrent neural network to learn quantum dynamics.
+    :property="og:image": https://pennylane.ai/qml/_images/qgrnn_thumbnail.png
+
+*Author: Jack Ceroni. Posted: 27 July 2020. Last updated: 26 Oct 2020.*


Reminder to change the "last updated" to the day this gets merged in :)

demonstrations/tutorial_qgrnn.py

glassnotes · 2021-02-24T23:05:00Z

demonstrations/tutorial_qgrnn.py

+        z_term = x_term = 1
+        for j in range(0, n_qubits):
+            if j == i:
+                z_term = np.kron(z_term, qml.PauliZ.matrix)


Nice, renaming the variables in this part makes things much clearer!

demonstrations/tutorial_qgrnn.py

glassnotes · 2021-02-24T23:09:13Z

demonstrations/tutorial_qgrnn.py

+#
+
+print("Target parameters \tLearned parameters")
+print("\t Weights:")


Would remove the \t here and left-justify, then add a row of ------ and some spacing underneath to help with separating the two types of parameters. Something like this maybe:

Target parameters Learned parameters Weights ------------------------------------------------ 0.56 0.5782895244479018 1.24 1.335028329676283 1.67 1.804480439985849 -0.79 -0.8395497395039528 Bias ------------------------------------------------ -1.44 -1.352958693194635 -1.43 -1.313891802560214 1.18 0.9811173767093425 -0.93 -1.0579331212003393

Added a bar too "|"

Co-authored-by: Olivia Di Matteo <[email protected]>

…to qgrnn_updates

glassnotes

Just two small additional changes, but once those are done, go ahead and merge 🎉

glassnotes · 2021-02-25T16:28:49Z

demonstrations/tutorial_qgrnn.py

+    :property="og:description": Using a quantum graph recurrent neural network to learn quantum dynamics.
+    :property="og:image": https://pennylane.ai/qml/_images/qgrnn_thumbnail.png
+
+*Author: Jack Ceroni. Posted: 27 July 2020. Last updated: 25 March 2021.*


Suggested change

*Author: Jack Ceroni. Posted: 27 July 2020. Last updated: 25 March 2021.*

*Author: Jack Ceroni. Posted: 27 July 2020. Last updated: 25 Feb 2021.*

I have no idea how that happened...

Thinking ahead 😀

glassnotes · 2021-02-25T16:31:38Z

demonstrations/tutorial_qgrnn.py

+axes[2].set_title("Learned", y=1.13)
+
+plt.subplots_adjust(wspace=0.3, hspace=0.3)
+plt.show()


The rendered plot here is quite small, any way to increase it? (Sorry didn't catch this earlier, I wasn't sure if in the previous iteration it was the old static image or the rendered image that was small.)

This seems to be less an issue with the image and more the amount of padding given to the image. Is there a way to reduce the padding given to the generated figure? I'm not too familiar with the procedure used to generate the website.

I could also switch it to have "initial" in its own row and then "target" and "learned" sharing a row. That would widen each plot.

Hmm... maybe try increasing the figsize, and/or throwing a plt.tight_layout() in there rather than plt.subplots_adjust? I know it's a pain to test things out because the runtime is too long to iterate quickly. If you can't get something working after a try or two we should ask Josh since he's the most knowledgeable about the demo build process.

If you can't get something working after a try or two we should ask Josh since he's the most knowledgeable about the demo build process.

Unfortunately I don't have any good solutions here! Normally I would download the Jupyter notebook version of the demo (or just copy the demo into a jupyter notebook), run all the cells, and then that allows you to iterate on the final plotting without performing the optimization every time.

The matplotlib image created in the notebook should match exactly the one on the website; Sphinx does not do any post-processing (I believe).

Given any updates to the plot generating code will probably make the code more complicated, and any improvements to sphinx to display it better are out of our technical depth, I think at this point it may be just better to leave as is. I'm merging it in.

glassnotes · 2021-02-25T16:32:16Z

demonstrations.rst

@@ -100,7 +100,7 @@ Demos
                    </a>
                </li>
                <li>
-                    <a href="demos/qgrnn.html">
+                    <a href="demos/tutorial_qgrnn.html">


Nice catch!

albi3ro added 2 commits January 25, 2021 11:38

minor updates, separate weights and bias, change some variables names

e2ee6b9

clean

240f6b4

albi3ro added the review-ready label Jan 25, 2021

josh146 requested review from glassnotes and thisac January 26, 2021 04:09

josh146 removed request for thisac and glassnotes January 26, 2021 04:25

Merge branch 'master' into qgrnn_updates

fd4a2ca

albi3ro and others added 5 commits February 24, 2021 12:55

Merge branch 'master' into qgrnn_updates

e22d46b

qgrnn turned into a tutorial

e4579e5

minor fixes

39a8a2a

change link

28e6933

change link

6b0509d

albi3ro requested a review from glassnotes February 24, 2021 21:55

one more link

e6756dc

glassnotes reviewed Feb 24, 2021

View reviewed changes

albi3ro and others added 3 commits February 25, 2021 09:08

Apply suggestions from code review

4200502

Co-authored-by: Olivia Di Matteo <[email protected]>

format final printout

12c645e

Merge branch 'qgrnn_updates' of https://github.com/PennyLaneAI/qml in…

dc4b542

…to qgrnn_updates

albi3ro requested a review from glassnotes February 25, 2021 16:00

glassnotes approved these changes Feb 25, 2021

View reviewed changes

albi3ro added 2 commits March 4, 2021 09:24

Merge branch 'master' into qgrnn_updates

d0b0bf9

Merge branch 'master' into qgrnn_updates

157aef9

albi3ro merged commit 50627ce into master Mar 5, 2021

albi3ro deleted the qgrnn_updates branch March 5, 2021 16:15

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Qgrnn updates #207

Qgrnn updates #207

albi3ro commented Jan 25, 2021

josh146 commented Jan 26, 2021

albi3ro commented Jan 26, 2021

josh146 commented Jan 26, 2021

glassnotes commented Jan 26, 2021

albi3ro commented Jan 26, 2021

glassnotes commented Jan 26, 2021

glassnotes commented Feb 22, 2021

glassnotes left a comment

glassnotes Feb 24, 2021

glassnotes Feb 24, 2021

albi3ro Feb 25, 2021

glassnotes Feb 24, 2021

albi3ro Feb 25, 2021

glassnotes left a comment

glassnotes Feb 25, 2021

albi3ro Feb 25, 2021

glassnotes Feb 25, 2021

glassnotes Feb 25, 2021

albi3ro Feb 25, 2021

glassnotes Feb 25, 2021

josh146 Feb 28, 2021

albi3ro Mar 5, 2021

glassnotes Feb 25, 2021

	Author: Jack Ceroni. Posted: 27 July 2020. Last updated: 25 March 2021.
	Author: Jack Ceroni. Posted: 27 July 2020. Last updated: 25 Feb 2021.

Qgrnn updates #207

Qgrnn updates #207

Conversation

albi3ro commented Jan 25, 2021

josh146 commented Jan 26, 2021

albi3ro commented Jan 26, 2021

josh146 commented Jan 26, 2021

glassnotes commented Jan 26, 2021

albi3ro commented Jan 26, 2021

glassnotes commented Jan 26, 2021

glassnotes commented Feb 22, 2021

glassnotes left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

glassnotes left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment