def callback(env):
"""internal function"""
if not state:
init(env)
bst = env.model
i = env.iteration
cvfolds = env.cvfolds
res_dict = {}
if i % skip_every == 1:
return
##### evaluation #####
if cvfolds is not None:
for feval in fevals:
tmp = aggcv([f.eval(i, feval) for f in cvfolds])
for k, mean, std in tmp:
res_dict[k] = [mean, std]
else:
for feval in fevals:
bst_eval = bst.eval_set(evals, i, feval)
res = [x.split(":") for x in bst_eval.split()]
for kv in res[1:]:
res_dict[kv[0]] = [float(kv[1])]
eval_res = []
keys = list(res_dict.keys())
keys.sort(key=lambda x: x if metric_shortname not in x else "a" + x)
for key in keys:
v = res_dict[key]
eval_res.append([key] + v)
##### print eval result #####
if not isinstance(verbose_eval,
bool) and verbose_eval and i % verbose_eval == 0:
infos = ["XGB iter: %3d" % i]
for item in eval_res:
if "null" in item[0]:
continue
infos.append("%s: %.6f" % (item[0], item[1]))
logger.debug("\t".join(infos))
if log_file:
with open(log_file, "a") as fout:
fout.write("\t".join(infos) + "\n")
##### choose score and do early stopping #####
score = None
for item in eval_res:
if item[0] == metric:
score = item[1]
break
assert score is not None
best_score = state["best_score"]
best_iteration = state["best_iteration"]
maximize_score = state["maximize_score"]
if (maximize_score and score > best_score) or (not maximize_score
and score < best_score):
msg = "[%d] %s" % (env.iteration, "\t".join(
[_fmt_metric(x) for x in eval_res]))
state["best_msg"] = msg
state["best_score"] = score
state["best_iteration"] = env.iteration
# save the property to attributes, so they will occur in checkpoint.
if env.model is not None:
env.model.set_attr(
best_score=str(state["best_score"]),
best_iteration=str(state["best_iteration"]),
best_msg=state["best_msg"],
)
elif env.iteration - best_iteration >= stopping_rounds:
best_msg = state["best_msg"]
if verbose_eval and env.rank == 0:
logger.debug("XGB stopped. Best iteration: %s ", best_msg)
raise EarlyStopException(best_iteration)