def _run_seq_trials(batch_run, suggest_x_cb):
proto_flag_vals = batch_run.batch_proto.get("flags")
batch_flag_vals = batch_run.get("flags")
max_trials = batch_run.get("max_trials") or DEFAULT_MAX_TRIALS
random_state = batch_run.get("random_seed")
random_starts = min(batch_flag_vals.get("random-starts") or 0, max_trials)
objective_scalar, objective_negate = _objective_y_info(batch_run)
prev_trials_cb = lambda: batch_util.trial_results(batch_run,
[objective_scalar])
trials_count = 0
for trial_flag_vals, is_trial_random_start, prev_trials, x0 in _iter_seq_trials(
proto_flag_vals,
objective_negate,
max_trials,
random_state,
random_starts,
prev_trials_cb,
suggest_x_cb,
batch_flag_vals,
):
_log_seq_trial(
is_trial_random_start,
random_starts,
trials_count,
x0,
prev_trials,
objective_scalar,
)
batch_util.run_trial(batch_run, trial_flag_vals)
trials_count += 1
def _run_seq_trials(batch_run, suggest_x_cb):
proto_flag_vals = batch_run.batch_proto.get("flags")
batch_flag_vals = suggest_opts = batch_run.get("flags")
max_trials = batch_run.get("max_trials") or DEFAULT_MAX_TRIALS
names, dims, initial_x = _flag_dims_for_search(proto_flag_vals)
random_state = batch_run.get("random_seed")
random_starts = min(batch_flag_vals.get("random-starts") or 0, max_trials)
objective_spec, objective_scalar, objective_negate = _objective_y_info(
batch_run)
runs_count = 0
for _ in range(max_trials):
prev_trials = batch_util.trial_results(batch_run, [objective_scalar])
x0, y0 = _trials_xy_for_prev_trials(prev_trials, names,
objective_negate)
suggest_random_start = _suggest_random_start(x0, runs_count,
random_starts)
_log_seq_trial(
suggest_random_start,
random_starts,
runs_count,
x0,
prev_trials,
objective_spec,
)
suggested_x, random_state = _suggest_x(suggest_x_cb, dims, x0, y0,
suggest_random_start,
random_state, suggest_opts)
if runs_count == 0 and suggested_x:
_apply_initial_x(initial_x, suggested_x)
trial_flag_vals = _trial_flags_for_x(suggested_x, names,
proto_flag_vals)
batch_util.run_trial(batch_run, trial_flag_vals)
runs_count += 1
def _run_seq_trials(batch_run, suggest_x_cb):
proto_flag_vals = batch_run.batch_proto.get("flags")
batch_flag_vals = batch_run.get("flags")
max_trials = batch_run.get("max_trials") or DEFAULT_MAX_TRIALS
random_state = batch_run.get("random_seed")
random_starts = min(batch_flag_vals.get("random-starts") or 0, max_trials)
objective_scalar, objective_negate = _objective_y_info(batch_run)
prev_trials_model = (batch_flag_vals.get("prev-trials")
or batch_util.PREV_TRIALS_BATCH)
prev_trials_cb = lambda: batch_util.trial_results(
batch_run, [objective_scalar], prev_trials_model)
trials_count = 0
for trial_flag_vals, is_trial_random_start, prev_trials, x0 in _iter_seq_trials(
proto_flag_vals,
objective_negate,
max_trials,
random_state,
random_starts,
prev_trials_cb,
suggest_x_cb,
batch_flag_vals,
):
_log_seq_trial(
is_trial_random_start,
random_starts,
trials_count,
x0,
prev_trials,
objective_scalar,
)
trial_run = batch_util.init_trial_run(batch_run, trial_flag_vals)
try:
batch_util.start_trial_run(trial_run)
except SystemExit as e:
batch_util.handle_trial_system_exit(e, batch_run, trial_run)
else:
trials_count += 1