def missing_search_dim_error(flag_vals):
log.error(
"flags for batch (%s) do not contain any search dimensions\n"
"Try specifying a range for one or more flags as NAME=[MIN:MAX].",
op_util.flags_desc(flag_vals),
)
raise SystemExit(1)
def _check_state_dims(state):
if not state.dim_names:
log.error(
"flags for batch (%s) do not contain any search " "dimension - quitting",
op_util.flags_desc(state.proto_flags),
)
raise batch_util.StopBatch(error=True)
def __call__(self, trial, state):
_check_state_dims(state)
next_trial_flags = self.seq_trial_cb(trial, state)
for run in trial.batch.seq_trial_runs():
if next_trial_flags == run.get("flags"):
log.warning("optimizer repeated trial (%s) - using random",
op_util.flags_desc(next_trial_flags))
next_trial_flags = self._random_trial_flags(state)
return next_trial_flags, {}
def __call__(self):
runs = []
results = []
trials = self.gen_trials(self.flags, runs, **self.opts)
for trial_flags, trial_opts in trials:
print("Running %s (%s):" %
(self.op.__name__, op_util.flags_desc(trial_flags)))
run, result = _run(self.op, trial_flags, trial_opts)
runs.append(run)
results.append(result)
return runs, results
def _row_for_print_run(run, flags, labels, status):
from guild.commands import runs_impl
fmt_run = runs_impl.format_run(run)
row = {"opspec": fmt_run["op_desc"]}
if flags:
flag_vals = run.get("flags") or {}
row["flags"] = op_util.flags_desc(flag_vals, delim=" ")
if labels:
row["label"] = run.get("label", "")
if status:
row["status"] = run.status
return row
def __call__(self):
runs = []
results = []
prev_results_cb = lambda: (runs, results)
for trial in self.gen_trials(self.flag_vals, prev_results_cb,
**self.opts):
trial_flag_vals, trial_attrs = _split_gen_trial(trial)
print("Running %s (%s):" %
(self.op.__name__, op_util.flags_desc(trial_flag_vals)))
run, result = _run(self.op, trial_flag_vals, self.opts,
trial_attrs)
runs.append(run)
results.append(result)
return runs, results
def gen_trials(flags,
_runs,
max_trials=None,
random_seed=None,
label=None,
**kw):
"""Public interface for ipy."""
if kw:
log.warning("ignoring configuration %s", kw)
num_trials = max_trials or DEFAULT_TRIALS
dim_names, dims, _initials = skopt_util.flag_dims(flags)
if not dim_names:
log.error(
"flags for batch (%s) do not contain any search "
"dimension - quitting", op_util.flags_desc(flags))
raise batch_util.StopBatch(error=True)
trial_vals = _gen_trial_vals(dims, num_trials, random_seed)
trial_opts = {"label": label or "random"}
return [(_trial_flags(dim_names, dim_vals, flags), trial_opts)
for dim_vals in trial_vals]
def _trial_flags_desc(run):
flags = {
name: val
for name, val in (run.get("flags") or {}).items() if val is not None
}
return op_util.flags_desc(flags)
def _flags_desc(self):
return op_util.flags_desc(self.flags)
