Dev #186

FelixXu35 · 2023-08-24T12:00:56Z

I partially rewrote the portfolio optimisation tutorial into two new ones.

Mixers and cvar are discussed in these two tutorials, respectively.

They passed the grammar, format and sphinx check.

If the new tutorials work, I am going to delete the old one.

review-notebook-app · 2023-08-24T12:01:00Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

docs/source/tutorials/QUBO_problem.ipynb

refraction-ray · 2023-08-24T12:28:43Z

docs/source/tutorials/QUBO_problem.ipynb

@@ -0,0 +1,580 @@
+{


it is not directly comparable in this figure, since the loss definition (different alpha) are different?

Reply via ReviewNB

Yes that's right. How low the loss can reach is not comparable. But you may find that with smaller alpha, the QAOA converges faster.

Hang on. I suddenly realise that they are comparable. In the ideal scenario, where the best result can be measured with 100% probability, their loss/cost should be the same.

but except the perfect case, they are not comparable

docs/source/tutorials/portfolio_optimization.ipynb

refraction-ray · 2023-08-24T12:28:44Z

docs/source/tutorials/portfolio_optimization.ipynb

@@ -0,0 +1,481 @@
+{


comparison with the one without mixer (plain QAOA)?

Reply via ReviewNB

It is compared with the one using standard X mixer, which is shown in the convergence figure and probability table above. With XY mixer, the convergence procedure isn't improved. But the overlap (probability of finding the best/correct result) increases.

Maybe include both lines (X and XY) in the same plot so the reader can see the comparison?

refraction-ray · 2023-08-24T12:30:00Z

also please modify this file accordingly is the same PR: https://github.com/tencent-quantum-lab/tensorcircuit/blob/master/docs/source/tutorial.rst

codecov · 2023-08-24T12:34:51Z

Codecov Report

Merging #186 (31d47ad) into master (b96a1b1) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #186   +/-   ##
=======================================
  Coverage   75.50%   75.50%           
=======================================
  Files          67       67           
  Lines       10668    10668           
=======================================
  Hits         8055     8055           
  Misses       2613     2613

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

refraction-ray · 2023-08-25T12:50:45Z

please use all lower letters for the file name, qubo instead of QUBO

docs/source/tutorials/portfolio_optimization.ipynb

JAllcock · 2023-08-29T07:56:05Z

docs/source/tutorials/portfolio_optimization.ipynb

@@ -0,0 +1,485 @@
+{


typo: 'brutal force' -> 'brute force'

Might be good to remind the reader about what the cost function is, e.g. "We first use brute force search to calculate the cost $x^TQx$ for each $x\in\{0,1\}^n$, where $n$ is the number of stocks.

Reply via ReviewNB

JAllcock · 2023-08-29T07:56:05Z

docs/source/tutorials/portfolio_optimization.ipynb

@@ -0,0 +1,485 @@
+{


'trained for 1000 times' -> 'trained for 1000 iterations'

Small thing, but earlier we talk about the 'cost' function, now we call it a 'loss'; maybe use the same term for both?

Reply via ReviewNB

JAllcock · 2023-08-29T07:56:05Z

docs/source/tutorials/portfolio_optimization.ipynb

@@ -0,0 +1,485 @@
+{


typo: 'thereby ensuring consistency with the classical approach' -> 'which is consistent with what we found with classical brute force search.'

Reply via ReviewNB

JAllcock · 2023-08-29T08:00:32Z

docs/source/tutorials/portfolio_optimization.ipynb

@@ -0,0 +1,485 @@
+{


There seems to be a lot of overlap between this tutorial and the other one 'qubo_problem.ipynb'. Better to split it roughly as (i) basic portfolio optimisation with QAOA (ii) advanced QAOA for portfolio optimisation?

In the first one you just do QAOA with X mixer; in the second you add XY mixer and also CVaR?

Reply via ReviewNB

JAllcock · 2023-08-29T08:01:47Z

docs/source/tutorials/qubo_problem.ipynb

@@ -0,0 +1,580 @@
+{


seems to be problem with latex rendering the Ising Hamiltonian

Reply via ReviewNB

JAllcock · 2023-08-29T08:22:02Z

docs/source/tutorials/qubo_problem.ipynb

@@ -0,0 +1,580 @@
+{


typo: 'is generally considered NP hard' -> 'is NP hard'
typo: 'we wish to minimise the cost function int he form of' -> 'we wish to minimise a cost function of the form'

The term 'ansatz' is mentioned a few times but isn't really defined

typo: 'this cost fucntion' -> 'this cost function'
typo: 'Below, the code of solving QUBO problem is shown' -> 'Below, the code used to solve this QUBO problem is shown.'

Reply via ReviewNB

JAllcock · 2023-08-29T08:22:02Z

docs/source/tutorials/qubo_problem.ipynb

@@ -0,0 +1,580 @@
+{


typo: 'the we see below' -> we see below
typo: 'combination' -> 'combination' (although maybe better to say 'give the best solution'

Reply via ReviewNB

JAllcock · 2023-08-29T08:22:02Z

docs/source/tutorials/qubo_problem.ipynb

@@ -0,0 +1,580 @@
+{


maybe add link to CVaR paper?

Reply via ReviewNB

JAllcock · 2023-08-29T08:22:02Z

docs/source/tutorials/qubo_problem.ipynb

@@ -0,0 +1,580 @@
+{


typo: 'let's define a function brutally calculate the costs (classical methods)' -> 'let's define a function to classically brute-force calculate all possible combinations of stocks and their associated cost functions.'

Reply via ReviewNB

JAllcock · 2023-08-29T08:22:02Z

docs/source/tutorials/qubo_problem.ipynb

@@ -0,0 +1,580 @@
+{


indicate what the legend (1, 0.25, 0.1) mean in both graphs?

Reply via ReviewNB

refraction-ray · 2023-09-14T04:57:38Z

tensorcircuit/applications/optimization.py

@@ -318,9 +330,10 @@ def QUBO_QAOA_cvar(
    :param maxiter: The maximum number of iterations for the optimization. Default is 1000.
    :return: The optimized parameters for the ansatz circuit.
    """
+    tf.config.run_functions_eagerly(True)


I dont think this is necessary for tf 2.0?

If this sentence is deleted, there is going to be an error like:

OperatorNotAllowedInGraphError: Iterating over a symbolic `tf.Tensor` is not allowed: AutoGraph did convert this function. This might indicate you are trying to use an unsupported feature.

exactly, this is because you impl is not jittable as I suggested below. By enabling this, the jit is closed. so no more no less, jit is still not enabled

I see. Will fix it very soon.

refraction-ray · 2023-09-14T07:07:53Z

tensorcircuit/applications/optimization.py


    # Iterate over the sorted results and calculate CVaR.
    while sump < percent:
-        if round(sump + p[count], 6) >= percent:
+        if tf.math.round(sump + p[count], 6) >= percent:


concerned on whether this is really jittable, due to the if tensor pattern here, where tensor directly depends on the jittable input argument of the function

I think the correct way is to build a sign function depending on the threshold to filter the results with higher loss without using the if tensor pattern

Thanks for your advice. Let me give it a try.

FelixXu35 · 2023-09-14T12:03:52Z

I refactored cvar_value, cvar_from_circuit, and cvar_from_expectation.
Now jit=True saves 1/3 time.

refraction-ray · 2023-09-15T02:41:23Z

Now jit=True saves 1/3 time

This is subtle, I expect orders of magnitude acceleration with jit. The implementation looks correct to me.

JAllcock

Number of typos, and places where the explanation e.g. docstring can be made clearer.

JAllcock · 2023-09-19T05:19:38Z

docs/source/tutorials/QAOA_funcs.py

+# moved
+def QUBO_to_Ising(Q: List[list]) -> Tuple[List[list], list, float]:
+    """
+    Cnvert the Q matrix into a the indication of pauli terms, the corresponding weights, and the offset.


spelling: Cnvert... into a the indication...
spelling: pauli -> Pauli

docstring would benefit from a simple example

JAllcock · 2023-09-19T05:21:54Z

docs/source/tutorials/QAOA_funcs.py

+
+def Ising_loss(c: tc.Circuit, pauli_terms: List[list], weights: list) -> float:
+    """
+    computes the loss function for the Ising model based on a given quantum circuit,


spelling: computes -> Computes

docstring would benefit from an explanation of what the loss function is

JAllcock · 2023-09-19T05:24:27Z

docs/source/tutorials/QAOA_funcs.py

+    :param ansatz: The ansatz function to be used for the QAOA.
+    :param nlayers: The number of layers in the QAOA ansatz.
+    :param iterations: The number of iterations to run the optimization.
+    :param vvag (optional): A flag indicating whether to use vectorized variational adjoint gradient. Default is False.


vvag is 'vectorized value and gradient'

JAllcock · 2023-09-19T05:32:36Z

docs/source/tutorials/QAOA_funcs.py

+    Performs the QAOA on a given QUBO problem.
+
+    :param Q: The n-by-n square and symmetric Q-matrix representing the QUBO problem.
+    :param ansatz: The ansatz function to be used for the QAOA.


I think you should say :param ansatz should be supplied e.g. by using the QAOA_ansatz_for_Ising function imported from templates.blocks?

Is 'ansatz' the circuit or is it the cost function defined by the circuit? below you define loss_val_grad in terms of the ansatz.

JAllcock · 2023-09-19T05:32:48Z

docs/source/tutorials/QAOA_funcs.py

+    return params
+
+
+# calcelled


JAllcock · 2023-09-19T05:48:06Z

tensorcircuit/applications/optimization.py

+) -> float:
+    """
+    Directly calculate the Conditional Value at Risk (CVaR) from a circuit.
+    The CVaR depends on a bunch of measurements.


'depends on a bunch of measurements' sounds too informal. Can we express this more formally?

JAllcock · 2023-09-19T05:53:27Z

tensorcircuit/applications/optimization.py

+    return cvar_result
+
+
+def cvar_from_expectation(circuit, Q, alpha: float) -> float:


This seems like the main difference between this and cvar_from_circuit is that cvar_from_circuit uses empirical measurement results, while cvar_from_expectation classically computes the probabilities of all outcomes. I think we should mention this.

JAllcock · 2023-09-19T05:54:44Z

tensorcircuit/templates/blocks.py

+    :param nlayers: The number of layers in the QAOA ansatz.
+    :pauli_terms: A list of Pauli terms, where each term is represented as a list of 0/1 series.
+    :param weights: A list of weights corresponding to each Pauli term.
+    :param mixer: mixer type. The options are "X", "XY_ring", "XY_par_ring", "XY_full", and "QAMPA".


I think an explicit example would help in the docstring

JAllcock · 2023-09-19T05:55:53Z

docs/source/tutorials/QAOA_funcs.py

+Tensor = Any
+
+
+# moved


you moved it to conversions right? why is it still here?

JAllcock · 2023-09-19T05:57:05Z

tensorcircuit/templates/conversions.py

+
+def QUBO_to_Ising(Q: List[list]) -> Tuple[List[list], list, float]:
+    """
+    Cnvert the Q matrix into a the indication of pauli terms, the corresponding weights, and the offset.


typo: Cnvert
typo: into a the indication of
typo: Pauli should be capitalized

refraction-ray

LGTM, thanks for the nice contribution!

FelixXu35 added 7 commits July 12, 2023 12:49

update the tutorials

cb82af6

add functions relavent of QAOA

4478367

Delete QAOA_funcs.py

98f7b39

clear the imports

e3a9eda

fixed minor grammar and format issues

5f4202f

Merge remote-tracking branch 'origin/ready-to-PR' into dev

ee10258

reverse two previously discarded changes

6c0fbc7

$refraction-ray$

refraction-ray reviewed Aug 24, 2023

View reviewed changes

reviewed V2

85e9f6a

FelixXu35 added 2 commits August 25, 2023 14:01

Merge branch 'master' into dev

e0ef92d

Rename QUBO_problem.ipynb to qubo_problem.ipynb

57d94bc

JAllcock reviewed Aug 29, 2023

View reviewed changes

FelixXu35 added 3 commits September 13, 2023 14:27

CVaR is now jittable

b14c67d

modify the qubo tutorials

2fa179b

Merge branch 'master' into dev

cf3ef90

$refraction-ray$

refraction-ray reviewed Sep 14, 2023

View reviewed changes

truelly jittable cvar

6a12efb

optimized cvar

31d47ad

JAllcock reviewed Sep 19, 2023

View reviewed changes

$refraction-ray$

refraction-ray approved these changes Sep 25, 2023

View reviewed changes

$@refraction-ray$ refraction-ray merged commit 31d47ad into tencent-quantum-lab:master Sep 25, 2023
4 checks passed

		return cvar_result


		def cvar_from_expectation(circuit, Q, alpha: float) -> float:

Dev #186

Dev #186

Conversation

FelixXu35 commented Aug 24, 2023

review-notebook-app bot commented Aug 24, 2023

refraction-ray Aug 24, 2023 • edited Loading

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

refraction-ray Aug 24, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

refraction-ray commented Aug 24, 2023

codecov bot commented Aug 24, 2023 • edited Loading

Codecov Report

refraction-ray commented Aug 25, 2023

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

Choose a reason for hiding this comment

JAllcock Aug 29, 2023 • edited Loading

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

FelixXu35 commented Sep 14, 2023

refraction-ray commented Sep 15, 2023 • edited Loading

JAllcock 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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

refraction-ray left a comment

Choose a reason for hiding this comment

$@refraction-ray$ refraction-ray Aug 24, 2023 •

edited

Loading

$@refraction-ray$ refraction-ray Aug 24, 2023 •

edited

Loading

codecov bot commented Aug 24, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

JAllcock Aug 29, 2023 •

edited

Loading

refraction-ray commented Sep 15, 2023 •

edited

Loading

$@refraction-ray$ refraction-ray left a comment