gdikov / hypertunity Goto Github PK

Hypertunity is a lightweight, high-level library for hyperparameter optimisation. Among others, it supports:

• Bayesian optimisation by wrapping GPyOpt,
• external or internal objective function evaluation by a scheduler, also compatible with Slurm,
• real-time visualisation of results in Tensorboard via the HParams plugin.

For the full set of features refer to the documentation.

Define the objective function to optimise. For example, it can take the hyperparameters params as input and return a raw value score as output:

import hypertunity as ht

def foo(**params) -> float:
    # do some very costly computations
    ...
    return score

To define the valid ranges for the values of params we create a Domain object:

domain = ht.Domain({
    "x": [-10., 10.],         # continuous variable within the interval [-10., 10.]
    "y": {"opt1", "opt2"},    # categorical variable from the set {"opt1", "opt2"}
    "z": set(range(4))        # discrete variable from the set {0, 1, 2, 3}
})

Then we set up the optimiser:

bo = ht.BayesianOptimisation(domain=domain)

And we run the optimisation for 10 steps. Each result is used to update the optimiser so that informed domain samples are drawn:

n_steps = 10
for i in range(n_steps):
    samples = bo.run_step(batch_size=2, minimise=True)      # suggest next samples
    evaluations = [foo(**s.as_dict()) for s in samples]     # evaluate foo
    bo.update(samples, evaluations)                         # update the optimiser

Finally, we visualise the results in Tensorboard:

import hypertunity.reports.tensorboard as tb

results = tb.Tensorboard(domain=domain, metrics=["score"], logdir="path/to/logdir")
results.from_history(bo.history)

A high-level wrapper class Trial allows for seamless parallel optimisation without bothering with scheduling jobs, updating optimisers and logging:

trial = ht.Trial(objective=foo,
                 domain=domain,
                 optimiser="bo",
                 reporter="tensorboard",
                 metrics=["score"])
trial.run(n_steps, batch_size=2, n_parallel=2)

To install the base version run:

pip install hypertunity

To use the Tensorboard dashboard, build the docs or run the test suite you will need the following extras:

pip install hypertunity[tensorboard,docs,tests]

Checkout the latest master and install locally:

git clone https://github.com/gdikov/hypertunity.git
cd hypertunity
pip install ./[tensorboard]