Feature/Replace qc.Loop with for loop using iPython magic #723

nulinspiratie · 2017-09-05T03:50:11Z

This PR adds the possibility to create QCoDeS loops using the Python for loop structure, which makes writing a Loop feel more Pythonic.
This is achieved using python cell magic: a measurement cell should be started with %%measurement (see below for an example).
The cell magic internally converts the for loop in a cell to the qc.Loop construct before executing the code.
Since iPython magic is used, it unfortunately only works in iPython (easiest in Jupyter Notebook).
The magic command has additional options, such as printing the executed code.

Example of new looping method

Old method to create a loop

Creating a Loop typically starts with creating a Loop object using qc.Loop.
After this, a dataset is created with the measurement_name

loop = qcodes.Loop({sweep_vals}).each(
    {measure_parameter1},
    {measure_parameter2},
    qcodes.Loop({sweep_vals2}).each(
        {measure_parameter3}))
data = loop.get_data_set(name={measurement_name})

{Additional code}

Using for loops

This can be replaced with for loops.
The first line should start with %%measurement followed by the measurement_name (and possible options).
After this, the Loop can be created by a for loop. Loops can also be nested same as you would nest for loops.

%%measurement {-options} {measurement_name}
for {sweep_vals}:
    {measure_parameter1}
    {measure_parameter2}
    for {sweep_vals2}:
        {measure_parameter3}

{Additional code}

Additional changes

The PR contains some additional changes related to magic. These are mainly to create a construct such that any future magic can be nicely integrated into QCoDeS

Instead of simply registering the %%measurement magic, it is contained in the magics class QCoDeSMagic. This is done to have access to additional magic-related features, such as parsing. It also makes it easier to stop a specific magic from being registered
The function register_magic_class is added, which handles the registering of a Magics class. The kwarg magic_commands allows only specific magics to be registered.
The item register_magic is added to config.core. It can be set to True, False, or a list (True by default).
The QCoDeS init checks if register_magic is not set to False and if the kernel is iPython. If so, then it will register magics. If register_magic=True, it will register all the magics in QCoDeSMagic. If register_magic is a list, it will only register those that are in the list.

@giulioungaretti @jenshnielsen @WilliamHPNielsen @core

… line magic

jenshnielsen · 2017-09-08T14:12:46Z

Cool

Can we have an explicit example? I assumed that the code you were meant to write would be
regular python code for i in {some generator} I had to try a few times to understand how it works

The code is only python 3.6 compatible because of the fstrings. We will likely break python 3.5 support soon but in the mean time it would be good to wrap the import in a check for python 3.6 and up.

nulinspiratie · 2017-09-11T03:11:15Z

@jenshnielsen I almost copied the Python for loop structure, but I don't instantiate a variable (i in your example). Instead, the for loop is for {some generator}, the reason being that the variable i can't actually be used in a Loop.

As an explicit example, say we have initialized parameter p = qc.ManualParameter('test_parameter'). Writing a loop could then be:

%%measurement test_measurement
for p.sweep(0, 10, step=1):
    p

which would translate to

loop = qc.Loop(p.sweep(0, 10, step=1)).each(
    p)
data = loop.get_data_set('test_measurement')

I'll add a check for Python 3.6

jenshnielsen · 2017-09-11T07:51:47Z

Yes I have figured out how it works, but your docstring could be more clear. I think you will need to document it with an explicit example.
Currently it's confusing because you write Create qcodes.Loop measurement using Python for syntax using iPython magic. which is not strictly true as for p.sweep(0, 10, step=1): can never be valid python. The formal definition of a for loop is for_stmt: 'for' exprlist 'in' testlist ':' suite ['else' ':' suite]
see https://docs.python.org/3/reference/grammar.html A for loop is not valid python with out the exprlint 'in' part

nulinspiratie · 2017-09-14T08:05:38Z

@jenshnielsen Yeah good point, I should've made it clearer. I'd love if we could actually use the setpoints as variables in a loop, in which case it would make sense to follow the same syntax as python for loops. I'll update the docstring to make this more clear. Other than that any comments?

jenshnielsen · 2017-09-14T08:31:30Z

I think the current version is a great start. I am hoping that when I finally get back to the new dataset which will be more disentangled from the loop that we can use for loops in a better way. Perhaps replace the magic with a context manager?

Only one tiny nitpick detail: it's actually IPython not iPython since it's not an apple product :)

nulinspiratie · 2017-09-15T01:03:43Z

@jenshnielsen I'm all for disentangling the dataset from the loop construct! As for context manager, I'm not seeing how we can use it for this type of manipulation, from my understanding it is only capable of performing actions before/after a piece of code, not actually altering the code itself. Can you elaborate?

…

On Thu, 14 Sep 2017 at 18:31 Jens Hedegaard Nielsen < ***@***.***> wrote: I think the current version is a great start. I am hoping that when I finally get back to the new dataset which will be more disentangled from the loop that we can use for loops in a better way. Perhaps replace the magic with a context manager? Only one tiny nitpick detail ist actually IPython not iPython since it's not an apple product :) — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#723 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGgLHfcfAhq6masdXC0OZHLQjFCWlCEjks5siORigaJpZM4PMaTq> .

jenshnielsen · 2017-09-15T07:18:29Z

I agree, My thinking is more in the direction of actually running the for loops as they are but having a context manager that takes care of all the dataset setup and so on but this will require quite some work and I am not sure if its even possible at this stage

sohailc · 2017-09-15T07:48:26Z

I am new to this tread so please forgive me if I am being ignorant. For the loop structure, what about a design as follows: Have a "sweep" method on the Parameter class which returns a generator. For each time "next" is called on this generator, the Parameter is set to the next value. To clarify, here is come mock code: for voltage in device.gate.sweep(np.linspace(0, 1, 10)): current = device.source_drain_current() And in the sweep code we have: def sweep(iterable): for value in iterable: self._set(value) yield value The nice thing about this is that we have a Python for loop with all its associated flexibility. We can extend this concept to two-D and n-D looping construct quite easily as well. How about doing stuff like: for voltage1, voltage2 in qczip(device.gate1, device.gate2).sweep(gate1=np.linspace(0, 1, 10), gate2=np.linspace(-1, 1, 15)): current = divice.source_drain_current() Have a look at itertools for inspiration: https://docs.python.org/2/library/itertools.html We can build things like measure functions and classes on top of this design. Maybe we can talk about this more in depth later? Love to hear from you guys. Sohail

…

On Fri, Sep 15, 2017 at 9:18 AM, Jens Hedegaard Nielsen < ***@***.***> wrote: I agree, My thinking is more in the direction of actually running the for loops as they are but having a context manager that takes care of all the dataset setup and so on but this will require quite some work and I am not sure if its even possible at this stage — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#723 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJ-GjLgUxmJO7w-WKf-wjQ2TXFGKq4L1ks5siiTHgaJpZM4PMaTq> .

sohailc · 2017-09-15T10:44:43Z

Why do we want to couple QCodes so tightly with IPython notebooks? What if we do not want to use notebooks as a way of running our experiments (as is indeed the talk here in Delft, that is that we should switch to regular python scripts)?

WilliamHPNielsen · 2017-09-15T11:13:38Z

@sohailc, generally, we don't want this strong coupling. QCoDeS should be useable without a notebook, and that is also the current state of affairs. In Copenhagen some people use Spyder in the lab. I agree with you that QCoDeS core functionality (drivers, datasets, loops) should not depend on the notebooks, but I think it is fine if people make notebook extensions of QCoDeS (like widgets or this magic).

sohailc · 2017-09-15T11:26:49Z

That is true. But I was under the impression that we want to get rid of the current looping construct in QCodes (right?) and instead want to make something new. I got the impression that that new way of looping would be dependent on IPython magic. But yes, I agree its totally fine to make IPython shortcuts :-)

…

On Fri, Sep 15, 2017 at 1:13 PM, William H.P. Nielsen < ***@***.***> wrote: @sohailc <https://github.com/sohailc>, generally, we don't want this strong coupling. QCoDeS should be useable without a notebook, and that is also the current state of affairs. I agree with you that QCoDeS core functionality (drivers, datasets, loops) should not depend on the notebooks, but I think it is fine if people make notebook extensions of QCoDeS (like widgets). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#723 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJ-GjOqaZ9Kcj6axIBHRi44dremLA4MZks5silvjgaJpZM4PMaTq> .

jenshnielsen · 2017-09-15T11:28:07Z

yes the title of this pull request is slightly miss leading. Its not a replacement but an alternative

sohailc · 2017-09-15T11:59:29Z

Ok cool. Do you or anyone else have comments on my suggestion of an alternative looping framework with generators?

…

On Fri, Sep 15, 2017 at 1:28 PM, Jens Hedegaard Nielsen < ***@***.***> wrote: yes the title of this pull request is slightly miss leading. Its not a replacement but an alternative — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#723 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJ-GjFdZRQ9zLefvUMDEORbeDk1UVwswks5sil9IgaJpZM4PMaTq> .

jenshnielsen · 2017-09-15T12:19:49Z

Its the way we want to go and also my original suggestion. My point is that we need a smart way to ensure that the data is captured and stored correctly so that when you do a for loop in a measurement you dont have to write too much boilerplate code to capture the data and setup plotting and what not

sohailc · 2017-09-15T12:27:57Z

Cool, I guess we came up with a similar idea independently. Looking forward to working with you guys.

…

On Fri, Sep 15, 2017 at 2:19 PM, Jens Hedegaard Nielsen < ***@***.***> wrote: Its the way we want to go and also my original suggestion. My point is that we need a smart way to ensure that the data is captured and stored correctly so that when you do a for loop in a measurement you dont have to write too much boilerplate code to capture the data and setup plotting and what not — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#723 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJ-GjOpzQcasFzU0xSHJqDH7gxHx1Dv8ks5simtlgaJpZM4PMaTq> .

nulinspiratie · 2017-09-17T04:59:56Z

@sohailc being able to use iterators/generators with a simple for loop structure would definitely result in measurements feeling much more Pythonic. I'm wondering how this can be implemented though. It seems to me -that meta information about the measurements is needed, such as its structure, the loop dimensionality etc. This enables creating an accompanying dataset. Since a for loop is a statement, it cannot be converted to an object, and so I'm not seeing how meta information can be gathered from this. Any thoughts?

sohailc · 2017-09-18T08:45:55Z

Hi @nulinspiratie thank you for your interesting thoughts and comments.

I think this can be accomplished by the sweep generator yielding a stateful object instead of just measurement numbers. What about doing somethings like this:

(A class method on the Parameter class):

def sweep(self, iterable):
  for value in iterable:
    self.set(iterable):
    yield self.instrument

In this case instrument is a stateful object. If we want to make a measurement dataset, we could make a measurement class which can be different for different measurement types. This measurement class takes as argument a generator which it will unrole when calling "run". The init method of this class can accept as argument a list of properties or methods which will enable it to extract measurement data from the stateful object for each time the generator yields the next value. For instance:

class Measurement:
  def init(self, properties): # and other useful info about the measurement, like data format
     self._properties = properties

  def run(self, generator): 
     for instrument in generator(): 
        for property in self._properties: 
            value = instrument.get(property)
            .... # code to store the property and value at this time.

So the complete measurement looks like:

m = qc.Measurement(["gate1", "sd1"])
m.run(instrument.sweep(gate1=np.linspace(0, 1, 10)))

I know this looks a bit like what we have already, but i think what is better about this design is that the run method can except any generator. Consider for example this:

m.run((instrument.set(value) for value in [0, 1, 2]))

Assuming that the set method returns "self". Another interesting thing you could do is perform non-linear sweeps. For example, finding the resonance frequency by stepping the frequency and iteratively finding the maximum value with newtons method:

def find_max_frequency(instrument, init):
  max_estimate = init
  while(True): # yield values until we have found the max
    .... # code to update the estimate of the max by measuring a local derivative. Adaptive stepping!
  yield instrument

My idea's have not matured, but consider how we could "zip" such a generator with a linear sweep to, for instance, flexibly program a measurement which finds a resonance at each gate voltage (for example).

sohailc · 2017-09-18T08:53:45Z

Another thing I should have pointed out is that I intended for the measurement class to be subclassed for each measurement type. So we can have (for example) "ResonanceMeasurement" and "MicrowaveMeasurement". So it should be easy to subclass the measurement class for non-programmers.

If we try to make a one-size-fits all class, I think we will fail.

nulinspiratie · 2017-09-21T02:14:09Z

Hi @sohailc very interesting ideas about future measurements, and thanks for taking the time to explain them! I've got a couple of questions related to it, I was hoping you could clarify.

First of all, what exactly is this stateful instrument that a parameter sweep returns? Is it an actual qc.Instrument object? If so, would this force parameters to belong to an instrument? Because most of ours do not belong to instruments. It also seems that the instrument is in charge of performing get operations. on each of the properties. How would the instrument know about these properties? On first glance, it seems as though this would force a link between the parameters that are swept (setters) and those that are measured (getters), is this correct?

Please correct me if I'm wrong, but is the main issue that's trying to be solved that we want to be able to specify the sweeping values adaptively, and that these may depend on previous measurement results? If so, the added benefit of an instrument is unclear to me.
For adaptive sweeping, would it not be sufficient if loops could accept generators of undefined length, and for parameters to have custom sweeps, which would be generators?
For this adaptive sweeping to be able to depend on measurement results, the Measurement (currently qc.Loop) would have to pass the measured values along to the parameter before the next sweep generator call.

Related to this, why would the Measurement be sublassed? Would it not make sense to simply create Measurement instances, such as the qc.Loop. Each instance would perform a specific measurement, such as measuring the resonance frequency. These can then be combined for more complex measurements.

Please let me know if anything is unclear, looking forward to your thoughts!

nulinspiratie · 2017-09-21T02:16:01Z

@jenshnielsen @WilliamHPNielsen @sohailc As for the PR, is this something we want to include in QCoDeS, or not? If so, I'll update the docstring to include a more concrete example

jenshnielsen · 2017-09-21T08:08:17Z

Im happy to include it. In this form it's optional and I can see it's usefulness. In the future we may be able to do something similar in a more clever way

WilliamHPNielsen · 2017-10-16T12:00:56Z

@nulinspiratie @jenshnielsen Is this still relevant? I guess we're waiting for the docstring update.

nulinspiratie · 2017-10-16T23:23:46Z

@WilliamHPNielsen I think so, I still use it all the time. I forgot about the docstring though, I'll do it early next week

…

On Mon, Oct 16, 2017, 11:00 PM William H.P. Nielsen < ***@***.***> wrote: @nulinspiratie <https://github.com/nulinspiratie> @jenshnielsen <https://github.com/jenshnielsen> Is this still relevant? I guess we're waiting for the docstring update. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#723 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGgLHR0EvbzIagaNsyU3F6Mgbrd8L6m5ks5ss0V5gaJpZM4PMaTq> .

…e/for_loop_magic

nulinspiratie · 2017-10-25T08:50:05Z

@WilliamHPNielsen @jenshnielsen I updated the docstring to mention the differing syntax. If you guys approve, I think it's ready to be merged

nulinspiratie · 2017-12-13T07:04:53Z

@jenshnielsen @WilliamHPNielsen any verdict?

WilliamHPNielsen · 2017-12-13T08:58:29Z

I haven't found the time to try it out, but since this is a parallel feature that people can use or ignore, I think it would not hurt merging it. @jenshnielsen do you disagree with that?

jenshnielsen · 2017-12-13T08:58:52Z

I think it's ok to merge

codecov · 2017-12-13T09:05:07Z

Codecov Report

Merging #723 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #723   +/-   ##
=======================================
  Coverage   78.63%   78.63%           
=======================================
  Files          33       33           
  Lines        4536     4536           
=======================================
  Hits         3567     3567           
  Misses        969      969

WilliamHPNielsen

Let's go!

Author: Serwan Asaad <[email protected]> Feature/Replace qc.Loop with for loop using iPython magic (#723)

nulinspiratie and others added 11 commits September 5, 2017 13:04

Added magic file

b274b8a

feat: add register_magic to config

bbdebbf

feat: add magic to qcodes init, including config

1c6f296

fix: change measurement to line_cell_magic, display help when used as…

0381c8c

… line magic

refactor: please the Codacy gods

6990836

fix: cleanup iPython check

e9776fc

fix: back to try: except

32a4aa5

fix: forgot a loop reference

2eefa43

feat: add delay functionality

4dc0ad4

fix: for_opts should be for_code

856b490

Merge branch 'master' into feature/for_loop_magic

7cb0e7e

feat: add check for python>=3.6

0878bc8

nulinspiratie added 2 commits October 25, 2017 19:48

doc: add mention of modified for syntax

249c0fb

Merge branch 'master' of https://github.com/QCoDeS/Qcodes into featur…

5b8a9e2

…e/for_loop_magic

Merge branch 'master' into feature/for_loop_magic

1d2a123

WilliamHPNielsen approved these changes Dec 13, 2017

View reviewed changes

jenshnielsen merged commit a7dbd3a into microsoft:master Dec 13, 2017

giulioungaretti pushed a commit that referenced this pull request Dec 13, 2017

Generated gh-pages for commit a7dbd3a

8e117a1

Author: Serwan Asaad <[email protected]> Feature/Replace qc.Loop with for loop using iPython magic (#723)

giulioungaretti pushed a commit that referenced this pull request Dec 13, 2017

Generated gh-pages for commit a7dbd3a

d4dc8fd

Author: Serwan Asaad <[email protected]> Feature/Replace qc.Loop with for loop using iPython magic (#723)

WilliamHPNielsen mentioned this pull request Dec 14, 2017

remove f-strings #916

Merged

nulinspiratie deleted the feature/for_loop_magic branch March 25, 2020 22:13

Feature/Replace qc.Loop with for loop using iPython magic #723

Feature/Replace qc.Loop with for loop using iPython magic #723

Uh oh!

Conversation

nulinspiratie commented Sep 5, 2017

Example of new looping method

Old method to create a loop

Using for loops

Additional changes

Uh oh!

jenshnielsen commented Sep 8, 2017

Uh oh!

nulinspiratie commented Sep 11, 2017

Uh oh!

jenshnielsen commented Sep 11, 2017

Uh oh!

nulinspiratie commented Sep 14, 2017

Uh oh!

jenshnielsen commented Sep 14, 2017 • edited by WilliamHPNielsen Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nulinspiratie commented Sep 15, 2017 via email

Uh oh!

jenshnielsen commented Sep 15, 2017

Uh oh!

sohailc commented Sep 15, 2017 via email • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sohailc commented Sep 15, 2017

Uh oh!

WilliamHPNielsen commented Sep 15, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sohailc commented Sep 15, 2017 via email

Uh oh!

jenshnielsen commented Sep 15, 2017

Uh oh!

sohailc commented Sep 15, 2017 via email

Uh oh!

jenshnielsen commented Sep 15, 2017

Uh oh!

sohailc commented Sep 15, 2017 via email

Uh oh!

nulinspiratie commented Sep 17, 2017

Uh oh!

sohailc commented Sep 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sohailc commented Sep 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nulinspiratie commented Sep 21, 2017

Uh oh!

nulinspiratie commented Sep 21, 2017

Uh oh!

jenshnielsen commented Sep 21, 2017 • edited by WilliamHPNielsen Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

WilliamHPNielsen commented Oct 16, 2017

Uh oh!

nulinspiratie commented Oct 16, 2017 via email

Uh oh!

nulinspiratie commented Oct 25, 2017

Uh oh!

nulinspiratie commented Dec 13, 2017

Uh oh!

WilliamHPNielsen commented Dec 13, 2017

Uh oh!

jenshnielsen commented Dec 13, 2017

Uh oh!

codecov bot commented Dec 13, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

WilliamHPNielsen left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

jenshnielsen commented Sep 14, 2017 •

edited by WilliamHPNielsen

Loading

sohailc commented Sep 15, 2017 via email •

edited

Loading

WilliamHPNielsen commented Sep 15, 2017 •

edited

Loading

sohailc commented Sep 18, 2017 •

edited

Loading

sohailc commented Sep 18, 2017 •

edited

Loading

jenshnielsen commented Sep 21, 2017 •

edited by WilliamHPNielsen

Loading

codecov bot commented Dec 13, 2017 •

edited

Loading