19.5. Asynchronous Successive Halving
Open the notebook in Colab
Open the notebook in Colab
Open the notebook in Colab
Open the notebook in Colab
Open the notebook in SageMaker Studio Lab

As we have seen in Section 19.3, we can accelerate HPO by distributing the evaluation of hyperparameter configurations across either multiple instances or multiples CPUs / GPUs on a single instance. However, compared to random search, it is not straightforward to run successive halving (SH) asynchronously in a distributed setting. Before we can decide which configuration to run next, we first have to collect all observations at the current rung level. This requires to synchronize workers at each rung level. For example, for the lowest rung level \(r_{\mathrm{min}}\), we first have to evaluate all \(N = \eta^K\) configurations, before we can promote the \(\frac{1}{\eta}\) of them to the next rung level.

In any distributed system, synchronization typically implies idle time for workers. First, we often observe high variations in training time across hyperparameter configurations. For example, assuming the number of filters per layer is a hyperparameter, then networks with less filters finish training faster than networks with more filters, which implies idle worker time due to stragglers. Moreover, the number of slots in a rung level is not always a multiple of the number of workers, in which case some workers may even sit idle for a full batch.

Figure Fig. 19.5.1 shows the scheduling of synchronous SH with \(\eta=2\) for four different trials with two workers. We start with evaluating Trial-0 and Trial-1 for one epoch and immediately continue with the next two trials once they are finished. We first have to wait until Trial-2 finishes, which takes substantially more time than the other trials, before we can promote the best two trials, i.e., Trial-0 and Trial-3 to the next rung level. This causes idle time for Worker-1. Then, we continue with Rung 1. Also, here Trial-3 takes longer than Trial-0, which leads to an additional ideling time of Worker-0. Once, we reach Rung-2, only the best trial, Trial-0, remains which occupies only one worker. To avoid that Worker-1 idles during that time, most implementaitons of SH continue already with the next round, and start evaluating new trials (e.g Trial-4) on the first rung.

../_images/sync_sh.svg

Fig. 19.5.1 Synchronous successive halving with two workers.

Asynchronous successive halving (ASHA) (Li et al., 2018) adapts SH to the asynchronous parallel scenario. The main idea of ASHA is to promote configurations to the next rung level as soon as we collected at least \(\eta\) observations on the current rung level. This decision rule may lead to suboptimal promotions: configurations can be promoted to the next rung level, which in hindsight do not compare favourably against most others at the same rung level. On the other hand, we get rid of all synchronization points this way. In practice, such suboptimal initial promotions have only a modest impact on performance, not only because the ranking of hyperparameter configurations is often fairly consistent across rung levels, but also because rungs grow over time and reflect the distribution of metric values at this level better and better. If a worker is free, but no configuration can be promoted, we start a new configuration with \(r = r_{\mathrm{min}}\), i.e the first rung level.

Fig. 19.5.2 shows the scheduling of the same configurations for ASHA. Once Trial-1 finishes, we collect the results of two trials (i.e Trial-0 and Trial-1) and immediately promote the better of them (Trial-0) to the next rung level. After Trial-0 finishes on rung 1, there are too few trials there in order to support a further promotion. Hence, we continue with rung 0 and evaluate Trial-3. Once Trial-3 finishes, Trial-2 is still pending. At this point we have 3 trials evaluated on rung 0 and one trial evaluated already on rung 1. Since Trial-3 performs worse than Trial-0 at rung 0, and \(\eta=2\), we cannot promote any new trial yet, and Worker-1 starts Trial-4 from scratch instead. However, once Trial-2 finishes and scores worse than Trial-3, the latter is promoted towards rung 1. Afterwards, we collected 2 evaluations on rung 1, which means we can now promote Trial-0 towards rung 2. At the same time, Worker-1 continues with evaluating new trials (i.e., Trial-5) on rung 0.

../_images/asha.svg

Fig. 19.5.2 Asynchronous successive halving (ASHA) with two workers.

import logging
from d2l import torch as d2l

logging.basicConfig(level=logging.INFO)
import matplotlib.pyplot as plt
from syne_tune import StoppingCriterion, Tuner
from syne_tune.backend.python_backend import PythonBackend
from syne_tune.config_space import loguniform, randint
from syne_tune.experiments import load_experiment
from syne_tune.optimizer.baselines import ASHA
INFO:root:SageMakerBackend is not imported since dependencies are missing. You can install them with
   pip install 'syne-tune[extra]'
AWS dependencies are not imported since dependencies are missing. You can install them with
   pip install 'syne-tune[aws]'
or (for everything)
   pip install 'syne-tune[extra]'
AWS dependencies are not imported since dependencies are missing. You can install them with
   pip install 'syne-tune[aws]'
or (for everything)
   pip install 'syne-tune[extra]'
INFO:root:Ray Tune schedulers and searchers are not imported since dependencies are missing. You can install them with
   pip install 'syne-tune[raytune]'
or (for everything)
   pip install 'syne-tune[extra]'

19.5.1. Objective Function

We will use Syne Tune with the same objective function as in Section 19.3.

def hpo_objective_lenet_synetune(learning_rate, batch_size, max_epochs):
    from syne_tune import Reporter
    from d2l import torch as d2l

    model = d2l.LeNet(lr=learning_rate, num_classes=10)
    trainer = d2l.HPOTrainer(max_epochs=1, num_gpus=1)
    data = d2l.FashionMNIST(batch_size=batch_size)
    model.apply_init([next(iter(data.get_dataloader(True)))[0]], d2l.init_cnn)
    report = Reporter()
    for epoch in range(1, max_epochs + 1):
        if epoch == 1:
            # Initialize the state of Trainer
            trainer.fit(model=model, data=data)
        else:
            trainer.fit_epoch()
        validation_error = trainer.validation_error().cpu().detach().numpy()
        report(epoch=epoch, validation_error=float(validation_error))

We will also use the same configuration space as before:

min_number_of_epochs = 2
max_number_of_epochs = 10
eta = 2

config_space = {
    "learning_rate": loguniform(1e-2, 1),
    "batch_size": randint(32, 256),
    "max_epochs": max_number_of_epochs,
}
initial_config = {
    "learning_rate": 0.1,
    "batch_size": 128,
}

19.5.2. Asynchronous Scheduler

First, we define the number of workers that evaluate trials concurrently. We also need to specify how long we want to run random search, by defining an upper limit on the total wall-clock time.

n_workers = 2  # Needs to be <= the number of available GPUs
max_wallclock_time = 12 * 60  # 12 minutes

The code for running ASHA is a simple variation of what we did for asynchronous random search.

mode = "min"
metric = "validation_error"
resource_attr = "epoch"

scheduler = ASHA(
    config_space,
    metric=metric,
    mode=mode,
    points_to_evaluate=[initial_config],
    max_resource_attr="max_epochs",
    resource_attr=resource_attr,
    grace_period=min_number_of_epochs,
    reduction_factor=eta,
)
INFO:syne_tune.optimizer.schedulers.fifo:max_resource_level = 10, as inferred from config_space
INFO:syne_tune.optimizer.schedulers.fifo:Master random_seed = 3140976097

Here, metric and resource_attr specify the key names used with the report callback, and max_resource_attr denotes which input to the objective function corresponds to \(r_{\mathrm{max}}\). Moreover, grace_period provides \(r_{\mathrm{min}}\), and reduction_factor is \(\eta\). We can run Syne Tune as before (this will take about 12 minutes):

trial_backend = PythonBackend(
    tune_function=hpo_objective_lenet_synetune,
    config_space=config_space,
)

stop_criterion = StoppingCriterion(max_wallclock_time=max_wallclock_time)
tuner = Tuner(
    trial_backend=trial_backend,
    scheduler=scheduler,
    stop_criterion=stop_criterion,
    n_workers=n_workers,
    print_update_interval=int(max_wallclock_time * 0.6),
)
tuner.run()
INFO:syne_tune.tuner:results of trials will be saved on /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046
INFO:root:Detected 4 GPUs
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.1 --batch_size 128 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/0/checkpoints
INFO:syne_tune.tuner:(trial 0) - scheduled config {'learning_rate': 0.1, 'batch_size': 128, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.44639554136672527 --batch_size 196 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/1/checkpoints
INFO:syne_tune.tuner:(trial 1) - scheduled config {'learning_rate': 0.44639554136672527, 'batch_size': 196, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.011548051321691994 --batch_size 254 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/2/checkpoints
INFO:syne_tune.tuner:(trial 2) - scheduled config {'learning_rate': 0.011548051321691994, 'batch_size': 254, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.14942487313193167 --batch_size 132 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/3/checkpoints
INFO:syne_tune.tuner:(trial 3) - scheduled config {'learning_rate': 0.14942487313193167, 'batch_size': 132, 'max_epochs': 10}
INFO:syne_tune.tuner:Trial trial_id 1 completed.
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.06317157191455719 --batch_size 242 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/4/checkpoints
INFO:syne_tune.tuner:(trial 4) - scheduled config {'learning_rate': 0.06317157191455719, 'batch_size': 242, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.48801815412811467 --batch_size 41 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/5/checkpoints
INFO:syne_tune.tuner:(trial 5) - scheduled config {'learning_rate': 0.48801815412811467, 'batch_size': 41, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.5904067586747807 --batch_size 244 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/6/checkpoints
INFO:syne_tune.tuner:(trial 6) - scheduled config {'learning_rate': 0.5904067586747807, 'batch_size': 244, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.08812857364095393 --batch_size 148 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/7/checkpoints
INFO:syne_tune.tuner:(trial 7) - scheduled config {'learning_rate': 0.08812857364095393, 'batch_size': 148, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.012271314788363914 --batch_size 235 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/8/checkpoints
INFO:syne_tune.tuner:(trial 8) - scheduled config {'learning_rate': 0.012271314788363914, 'batch_size': 235, 'max_epochs': 10}
INFO:syne_tune.tuner:Trial trial_id 5 completed.
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.08845692598296777 --batch_size 236 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/9/checkpoints
INFO:syne_tune.tuner:(trial 9) - scheduled config {'learning_rate': 0.08845692598296777, 'batch_size': 236, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.0825770880068151 --batch_size 75 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/10/checkpoints
INFO:syne_tune.tuner:(trial 10) - scheduled config {'learning_rate': 0.0825770880068151, 'batch_size': 75, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.20235201406823256 --batch_size 65 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/11/checkpoints
INFO:syne_tune.tuner:(trial 11) - scheduled config {'learning_rate': 0.20235201406823256, 'batch_size': 65, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.3359885631737537 --batch_size 58 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/12/checkpoints
INFO:syne_tune.tuner:(trial 12) - scheduled config {'learning_rate': 0.3359885631737537, 'batch_size': 58, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.7892434579795236 --batch_size 89 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/13/checkpoints
INFO:syne_tune.tuner:(trial 13) - scheduled config {'learning_rate': 0.7892434579795236, 'batch_size': 89, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.1233786579597858 --batch_size 176 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/14/checkpoints
INFO:syne_tune.tuner:(trial 14) - scheduled config {'learning_rate': 0.1233786579597858, 'batch_size': 176, 'max_epochs': 10}
INFO:syne_tune.tuner:Trial trial_id 13 completed.
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.13707981127012328 --batch_size 141 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/15/checkpoints
INFO:syne_tune.tuner:(trial 15) - scheduled config {'learning_rate': 0.13707981127012328, 'batch_size': 141, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.02913976299993913 --batch_size 116 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/16/checkpoints
INFO:syne_tune.tuner:(trial 16) - scheduled config {'learning_rate': 0.02913976299993913, 'batch_size': 116, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.033362897489792855 --batch_size 154 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/17/checkpoints
INFO:syne_tune.tuner:(trial 17) - scheduled config {'learning_rate': 0.033362897489792855, 'batch_size': 154, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.29442952580755816 --batch_size 210 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/18/checkpoints
INFO:syne_tune.tuner:(trial 18) - scheduled config {'learning_rate': 0.29442952580755816, 'batch_size': 210, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.10214259921521483 --batch_size 239 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/19/checkpoints
INFO:syne_tune.tuner:(trial 19) - scheduled config {'learning_rate': 0.10214259921521483, 'batch_size': 239, 'max_epochs': 10}
INFO:syne_tune.tuner:tuning status (last metric is reported)
 trial_id     status  iter  learning_rate  batch_size  max_epochs  epoch  validation_error  worker-time
        0    Stopped     4       0.100000         128          10    4.0          0.430578    29.093798
        1  Completed    10       0.446396         196          10   10.0          0.205652    72.747496
        2    Stopped     2       0.011548         254          10    2.0          0.900570    13.729115
        3    Stopped     8       0.149425         132          10    8.0          0.259171    58.980305
        4    Stopped     4       0.063172         242          10    4.0          0.900579    27.773950
        5  Completed    10       0.488018          41          10   10.0          0.140488   113.171314
        6    Stopped    10       0.590407         244          10   10.0          0.193776    70.364757
        7    Stopped     2       0.088129         148          10    2.0          0.899955    14.169738
        8    Stopped     2       0.012271         235          10    2.0          0.899840    13.434274
        9    Stopped     2       0.088457         236          10    2.0          0.899801    13.034437
       10    Stopped     4       0.082577          75          10    4.0          0.385970    35.426524
       11    Stopped     4       0.202352          65          10    4.0          0.543102    34.653495
       12    Stopped    10       0.335989          58          10   10.0          0.149558    90.924182
       13  Completed    10       0.789243          89          10   10.0          0.144887    77.365970
       14    Stopped     2       0.123379         176          10    2.0          0.899987    12.422906
       15    Stopped     2       0.137080         141          10    2.0          0.899983    13.395153
       16    Stopped     4       0.029140         116          10    4.0          0.900532    27.834111
       17    Stopped     2       0.033363         154          10    2.0          0.899996    13.407285
       18 InProgress     1       0.294430         210          10    1.0          0.899878     6.126259
       19 InProgress     0       0.102143         239          10      -                 -            -
2 trials running, 18 finished (3 until the end), 437.07s wallclock-time

INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.02846298236356246 --batch_size 115 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/20/checkpoints
INFO:syne_tune.tuner:(trial 20) - scheduled config {'learning_rate': 0.02846298236356246, 'batch_size': 115, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.037703019195187606 --batch_size 91 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/21/checkpoints
INFO:syne_tune.tuner:(trial 21) - scheduled config {'learning_rate': 0.037703019195187606, 'batch_size': 91, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.0741039859356903 --batch_size 192 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/22/checkpoints
INFO:syne_tune.tuner:(trial 22) - scheduled config {'learning_rate': 0.0741039859356903, 'batch_size': 192, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.3032613031191755 --batch_size 252 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/23/checkpoints
INFO:syne_tune.tuner:(trial 23) - scheduled config {'learning_rate': 0.3032613031191755, 'batch_size': 252, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.019823425532533637 --batch_size 252 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/24/checkpoints
INFO:syne_tune.tuner:(trial 24) - scheduled config {'learning_rate': 0.019823425532533637, 'batch_size': 252, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.8203370335228594 --batch_size 77 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/25/checkpoints
INFO:syne_tune.tuner:(trial 25) - scheduled config {'learning_rate': 0.8203370335228594, 'batch_size': 77, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.2960420911378594 --batch_size 104 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/26/checkpoints
INFO:syne_tune.tuner:(trial 26) - scheduled config {'learning_rate': 0.2960420911378594, 'batch_size': 104, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.2993874715754653 --batch_size 192 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/27/checkpoints
INFO:syne_tune.tuner:(trial 27) - scheduled config {'learning_rate': 0.2993874715754653, 'batch_size': 192, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.08056711961080017 --batch_size 36 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/28/checkpoints
INFO:syne_tune.tuner:(trial 28) - scheduled config {'learning_rate': 0.08056711961080017, 'batch_size': 36, 'max_epochs': 10}
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.26868380288030347 --batch_size 151 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/29/checkpoints
INFO:syne_tune.tuner:(trial 29) - scheduled config {'learning_rate': 0.26868380288030347, 'batch_size': 151, 'max_epochs': 10}
INFO:syne_tune.tuner:Trial trial_id 29 completed.
INFO:root:running subprocess with command: /usr/bin/python /home/ci/.local/lib/python3.8/site-packages/syne_tune/backend/python_backend/python_entrypoint.py --learning_rate 0.9197404791177789 --batch_size 66 --max_epochs 10 --tune_function_root /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/tune_function --tune_function_hash e03d187e043d2a17cae636d6af164015 --st_checkpoint_dir /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046/30/checkpoints
INFO:syne_tune.tuner:(trial 30) - scheduled config {'learning_rate': 0.9197404791177789, 'batch_size': 66, 'max_epochs': 10}
INFO:syne_tune.stopping_criterion:reaching max wallclock time (720), stopping there.
INFO:syne_tune.tuner:Stopping trials that may still be running.
INFO:syne_tune.tuner:Tuning finished, results of trials can be found on /home/ci/syne-tune/python-entrypoint-2023-08-18-20-01-52-046
--------------------
Resource summary (last result is reported):
 trial_id     status  iter  learning_rate  batch_size  max_epochs  epoch  validation_error  worker-time
        0    Stopped     4       0.100000         128          10      4          0.430578    29.093798
        1  Completed    10       0.446396         196          10     10          0.205652    72.747496
        2    Stopped     2       0.011548         254          10      2          0.900570    13.729115
        3    Stopped     8       0.149425         132          10      8          0.259171    58.980305
        4    Stopped     4       0.063172         242          10      4          0.900579    27.773950
        5  Completed    10       0.488018          41          10     10          0.140488   113.171314
        6    Stopped    10       0.590407         244          10     10          0.193776    70.364757
        7    Stopped     2       0.088129         148          10      2          0.899955    14.169738
        8    Stopped     2       0.012271         235          10      2          0.899840    13.434274
        9    Stopped     2       0.088457         236          10      2          0.899801    13.034437
       10    Stopped     4       0.082577          75          10      4          0.385970    35.426524
       11    Stopped     4       0.202352          65          10      4          0.543102    34.653495
       12    Stopped    10       0.335989          58          10     10          0.149558    90.924182
       13  Completed    10       0.789243          89          10     10          0.144887    77.365970
       14    Stopped     2       0.123379         176          10      2          0.899987    12.422906
       15    Stopped     2       0.137080         141          10      2          0.899983    13.395153
       16    Stopped     4       0.029140         116          10      4          0.900532    27.834111
       17    Stopped     2       0.033363         154          10      2          0.899996    13.407285
       18    Stopped     8       0.294430         210          10      8          0.241193    52.089688
       19    Stopped     2       0.102143         239          10      2          0.900002    12.487762
       20    Stopped     2       0.028463         115          10      2          0.899995    14.100359
       21    Stopped     2       0.037703          91          10      2          0.900026    14.664848
       22    Stopped     2       0.074104         192          10      2          0.901730    13.312770
       23    Stopped     2       0.303261         252          10      2          0.900009    12.725821
       24    Stopped     2       0.019823         252          10      2          0.899917    12.533380
       25    Stopped    10       0.820337          77          10     10          0.196842    81.816103
       26    Stopped    10       0.296042         104          10     10          0.198453    81.121330
       27    Stopped     4       0.299387         192          10      4          0.336183    24.610689
       28 InProgress     9       0.080567          36          10      9          0.203052   104.303746
       29  Completed    10       0.268684         151          10     10          0.222814    68.217289
       30 InProgress     1       0.919740          66          10      1          0.900037    10.070776
2 trials running, 29 finished (4 until the end), 723.70s wallclock-time

validation_error: best 0.1404876708984375 for trial-id 5
--------------------

Note that we are running a variant of ASHA where underperforming trials are stopped early. This is different to our implementation in Section 19.4.1, where each training job is started with a fixed max_epochs. In the latter case, a well-performing trial which reaches the full 10 epochs, first needs to train 1, then 2, then 4, then 8 epochs, each time starting from scratch. This type of pause-and-resume scheduling can be implemented efficiently by checkpointing the training state after each epoch, but we avoid this extra complexity here. After the experiment has finished, we can retrieve and plot results.

d2l.set_figsize()
e = load_experiment(tuner.name)
e.plot()
WARNING:matplotlib.legend:No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.
../_images/output_sh-async_bb0ea6_13_1.svg

19.5.3. Visualize the Optimization Process

Once more, we visualize the learning curves of every trial (each color in the plot represents a trial). Compare this to asynchronous random search in Section 19.3. As we have seen for successive halving in Section 19.4, most of the trials are stopped at 1 or 2 epochs (\(r_{\mathrm{min}}\) or \(\eta * r_{\mathrm{min}}\)). However, trials do not stop at the same point, because they require different amount of time per epoch. If we ran standard successive halving instead of ASHA, we would need to synchronize our workers, before we can promote configurations to the next rung level.

d2l.set_figsize([6, 2.5])
results = e.results
for trial_id in results.trial_id.unique():
    df = results[results["trial_id"] == trial_id]
    d2l.plt.plot(
        df["st_tuner_time"],
        df["validation_error"],
        marker="o"
    )
d2l.plt.xlabel("wall-clock time")
d2l.plt.ylabel("objective function")
Text(0, 0.5, 'objective function')
../_images/output_sh-async_bb0ea6_15_1.svg

19.5.4. Summary

Compared to random search, successive halving is not quite as trivial to run in an asynchronous distributed setting. To avoid synchronisation points, we promote configurations as quickly as possible to the next rung level, even if this means promoting some wrong ones. In practice, this usually does not hurt much, and the gains of asynchronous versus synchronous scheduling are usually much higher than the loss of the suboptimal decision making.

Discussions