Advanced Topics#

Note

Because this documentation consists of static html, the live_plot and live_info widget is not live. Download the notebook in order to see the real behaviour. [1]

Saving and loading learners#

Every learner has a save and load method that can be used to save and load only the data of a learner.

Use the fname argument in learner.save(fname=...).

Or, when using a BalancingLearner one can use either a callable that takes the child learner and returns a filename or a list of filenames.

By default the resulting pickle files are compressed, to turn this off use learner.save(fname=..., compress=False)

# Let's create two learners and run only one.
learner = adaptive.Learner1D(f, bounds=(-1, 1))
control = adaptive.Learner1D(f, bounds=(-1, 1))

# Let's only run the learner
runner = adaptive.Runner(learner, loss_goal=0.01)

runner.live_info()

fname = "data/example_file.p"
learner.save(fname)
control.load(fname)

(learner.plot().relabel("saved learner") + control.plot().relabel("loaded learner"))

Or just (without saving):

control = adaptive.Learner1D(f, bounds=(-1, 1))
control.copy_from(learner)

One can also periodically save the learner while running in a Runner. Use it like:

def slow_f(x):
    from time import sleep

    sleep(5)
    return x


learner = adaptive.Learner1D(slow_f, bounds=[0, 1])
runner = adaptive.Runner(learner, npoints_goal=100)
runner.start_periodic_saving(
    save_kwargs={"fname": "data/periodic_example.p"}, interval=6
)

<Task pending name='Task-9' coro=<AsyncRunner.start_periodic_saving.<locals>._saver() running at /home/docs/checkouts/readthedocs.org/user_builds/adaptive/checkouts/latest/adaptive/runner.py:897>>

runner.live_info()  # we cancelled it after 6 seconds

# See the data 6 later seconds with
#!ls -lah data  # only works on macOS and Linux systems

A watched pot never boils!#

The adaptive.Runner does its work in an asyncio task that runs concurrently with the IPython kernel, when using adaptive from a Jupyter notebook. This is advantageous because it allows us to do things like live-updating plots, however it can trip you up if you’re not careful.

Notably: if you block the IPython kernel, the runner will not do any work.

For example if you wanted to wait for a runner to complete, do not wait in a busy loop:

while not runner.task.done():
    pass

If you do this then the runner will never finish.

What to do if you don’t care about live plotting, and just want to run something until its done?

The simplest way to accomplish this is to use adaptive.BlockingRunner:

learner = adaptive.Learner1D(f, bounds=(-1, 1))
adaptive.BlockingRunner(learner, loss_goal=0.01)
# This will only get run after the runner has finished
learner.plot()

Reproducibility#

By default adaptive runners evaluate the learned function in parallel across several cores. The runners are also opportunistic, in that as soon as a result is available they will feed it to the learner and request another point to replace the one that just finished.

Because the order in which computations complete is non-deterministic, this means that the runner behaves in a non-deterministic way. Adaptive makes this choice because in many cases the speedup from parallel execution is worth sacrificing the “purity” of exactly reproducible computations.

Nevertheless it is still possible to run a learner in a deterministic way with adaptive.

The simplest way is to use adaptive.runner.simple to run your learner:

learner = adaptive.Learner1D(f, bounds=(-1, 1))

# blocks until completion
adaptive.runner.simple(learner, loss_goal=0.01)

learner.plot()

Note that unlike adaptive.Runner, adaptive.runner.simple blocks until it is finished.

If you want to enable determinism, want to continue using the non-blocking adaptive.Runner, you can use the adaptive.runner.SequentialExecutor:

from adaptive.runner import SequentialExecutor

learner = adaptive.Learner1D(f, bounds=(-1, 1))
runner = adaptive.Runner(learner, executor=SequentialExecutor(), loss_goal=0.01)

runner.live_info()

runner.live_plot(update_interval=0.1)

Cancelling a runner#

Sometimes you want to interactively explore a parameter space, and want the function to be evaluated at finer and finer resolution and manually control when the calculation stops.

If no goal is provided to a runner then the runner will run until cancelled.

runner.live_info() will provide a button that can be clicked to stop the runner. You can also stop the runner programatically using runner.cancel().

learner = adaptive.Learner1D(f, bounds=(-1, 1))
runner = adaptive.Runner(learner)

/home/docs/checkouts/readthedocs.org/user_builds/adaptive/checkouts/latest/adaptive/runner.py:1251: UserWarning: Goal is None which means the learners continue forever!
  return auto_goal(

runner.cancel()  # Let's execute this after 0.1 seconds

runner.live_info()

runner.live_plot(update_interval=0.1)

print(runner.status())

cancelled

Debugging Problems#

Runners work in the background with respect to the IPython kernel, which makes it convenient, but also means that inspecting errors is more difficult because exceptions will not be raised directly in the notebook. Often the only indication you will have that something has gone wrong is that nothing will be happening.

Let’s look at the following example, where the function to be learned will raise an exception 10% of the time.

def will_raise(x):
    from random import random
    from time import sleep

    sleep(random())
    if random() < 0.1:
        raise RuntimeError("something went wrong!")
    return x**2


learner = adaptive.Learner1D(will_raise, (-1, 1))
runner = adaptive.Runner(
    learner
)  # without 'goal' the runner will run forever unless cancelled

/home/docs/checkouts/readthedocs.org/user_builds/adaptive/checkouts/latest/adaptive/runner.py:1251: UserWarning: Goal is None which means the learners continue forever!
  return auto_goal(

runner.live_info()

runner.live_plot()

The above runner should continue forever, but we notice that it stops after a few points are evaluated.

First we should check that the runner has really finished:

runner.task.done()

False

If it has indeed finished then we should check the result of the runner. This should be None if the runner stopped successfully. If the runner stopped due to an exception then asking for the result will raise the exception with the stack trace:

runner.task.result()

---------------------------------------------------------------------------
InvalidStateError                         Traceback (most recent call last)
Cell In[28], line 1
----> 1 runner.task.result()

InvalidStateError: Result is not set.

You can also check runner.tracebacks which is a list of tuples with (point, traceback).

for point, tb in runner.tracebacks:
    print(f"point: {point}:\n {tb}")

Logging runners#

Runners do their job in the background, which makes introspection quite cumbersome. One way to inspect runners is to instantiate one with log=True:

learner = adaptive.Learner1D(f, bounds=(-1, 1))
runner = adaptive.Runner(learner, loss_goal=0.01, log=True)

runner.live_info()

This gives a the runner a log attribute, which is a list of the learner methods that were called, as well as their arguments. This is useful because executors typically execute their tasks in a non-deterministic order.

This can be used with adaptive.runner.replay_log to perfom the same set of operations on another runner:

reconstructed_learner = adaptive.Learner1D(f, bounds=learner.bounds)
adaptive.runner.replay_log(reconstructed_learner, runner.log)

learner.plot().Scatter.I.opts(size=6) * reconstructed_learner.plot()

Adding coroutines#

In the following example we’ll add a Task that times the runner. This is only for demonstration purposes because one can simply check runner.elapsed_time() or use the runner.live_info() widget to see the time since the runner has started.

So let’s get on with the example. To time the runner you cannot simply use

now = datetime.now()
runner = adaptive.Runner(...)
print(datetime.now() - now)

because this will be done immediately. Also blocking the kernel with while not runner.task.done() will not work because the runner will not do anything when the kernel is blocked.

Therefore you need to create an async function and hook it into the ioloop like so:

import asyncio

async def time(runner):
    from datetime import datetime

    now = datetime.now()
    await runner.task
    return datetime.now() - now

ioloop = asyncio.get_event_loop()

learner = adaptive.Learner1D(f, bounds=(-1, 1))
runner = adaptive.Runner(learner, loss_goal=0.01)

timer = ioloop.create_task(time(runner))

# The result will only be set when the runner is done.
timer.result()

datetime.timedelta(microseconds=179854)

Custom parallelization using coroutines#

Adaptive by itself does not implement a way of sharing partial results between function executions. Instead its implementation of parallel computation using executors is minimal by design. The appropriate way to implement custom parallelization is by using coroutines (asynchronous functions).

We illustrate this approach by using dask.distributed for parallel computations in part because it supports asynchronous operation out-of-the-box. We will focus on a function f(x) that consists of two distinct components: a slow part g that can be reused across multiple inputs and shared among various function evaluations, and a fast part h that is calculated for each x value.

def f(x):  # example function without caching
    """
    Integer part of `x` repeats and should be reused
    Decimal part requires a new computation
    """
    return g(int(x)) + h(x % 1)


def g(x):
    """Slow but reusable function"""
    from time import sleep

    sleep(random.randrange(5))
    return x**2


def h(x):
    """Fast function"""
    return x**3

Using `adaptive.utils.daskify`#

To simplify the process of using coroutines and caching with dask and Adaptive, we provide the adaptive.utils.daskify() decorator. This decorator can be used to parallelize functions with caching as well as functions without caching, making it a powerful tool for custom parallelization in Adaptive.

from dask.distributed import Client

import adaptive

client = await Client(asynchronous=True)


# The g function has caching enabled
g_dask = adaptive.utils.daskify(client, cache=True)(g)

# Can be used like a decorator too:
# >>> @adaptive.utils.daskify(client, cache=True)
# ... def g(x): ...

# The h function does not use caching
h_dask = adaptive.utils.daskify(client)(h)

# Now we need to rewrite `f(x)` to use `g` and `h` as coroutines


async def f_parallel(x):
    g_result = await g_dask(int(x))
    h_result = await h_dask(x % 1)
    return (g_result + h_result) ** 2


learner = adaptive.Learner1D(f_parallel, bounds=(-3.5, 3.5))
runner = adaptive.AsyncRunner(learner, loss_goal=0.01, ntasks=20)
runner.live_info()

Finally, we wait for the runner to finish, and then plot the result.

await runner.task
learner.plot()

Step-by-step explanation of custom parallelization#

Now let’s dive into a detailed explanation of the process to understand how the adaptive.utils.daskify() decorator works.

In order to combine reuse of values of g with adaptive, we need to convert f into a dask graph by using dask.delayed.

from dask import delayed

# Convert g and h to dask.Delayed objects, such that they run in the Client
g, h = delayed(g), delayed(h)


@delayed
def f(x, y):
    return (x + y) ** 2

Next we define a computation using coroutines such that it reuses previously submitted tasks.

from dask.distributed import Client

client = await Client(asynchronous=True)

g_futures = {}


async def f_parallel(x):
    # Get or sumbit the slow function future
    if (g_future := g_futures.get(int(x))) is None:
        g_futures[int(x)] = g_future = client.compute(g(int(x)))

    future_f = client.compute(f(g_future, h(x % 1)))

    return await future_f

/home/docs/checkouts/readthedocs.org/user_builds/adaptive/conda/latest/lib/python3.10/site-packages/distributed/node.py:182: UserWarning: Port 8787 is already in use.
Perhaps you already have a cluster running?
Hosting the HTTP server on port 33319 instead
  warnings.warn(

To run the adaptive evaluation we provide the asynchronous function to the learner and run it via AsyncRunner without specifying an executor.

learner = adaptive.Learner1D(f_parallel, bounds=(-3.5, 3.5))

runner = adaptive.AsyncRunner(learner, loss_goal=0.01, ntasks=20)

Finally we wait for the runner to finish, and then plot the result.

await runner.task
learner.plot()

Using Runners from a script#

Runners can also be used from a Python script independently of the notebook.

The simplest way to accomplish this is simply to use the BlockingRunner:

import adaptive

def f(x):
    return x

learner = adaptive.Learner1D(f, (-1, 1))

adaptive.BlockingRunner(learner, loss_goal=0.1)

If you use asyncio already in your script and want to integrate adaptive into it, then you can use the default Runner as you would from a notebook. If you want to wait for the runner to finish, then you can simply

await runner.task

from within a coroutine.

Advanced Topics#

Saving and loading learners#

A watched pot never boils!#

Reproducibility#

Cancelling a runner#

Debugging Problems#

Logging runners#

Adding coroutines#

Custom parallelization using coroutines#

Using adaptive.utils.daskify#

Step-by-step explanation of custom parallelization#

Using Runners from a script#

Using `adaptive.utils.daskify`#