def compute_metrics(cols, y_true, y_score, num_tasks, cal_fact_aucpr):
if len(cols) < 1:
return pd.DataFrame(
{
"roc_auc_score": np.nan,
"auc_pr": np.nan,
"avg_prec_score": np.nan,
"f1_max": np.nan,
"p_f1_max": np.nan,
"kappa": np.nan,
"kappa_max": np.nan,
"p_kappa_max": np.nan,
"bceloss": np.nan
},
index=np.arange(num_tasks))
df = pd.DataFrame({"task": cols, "y_true": y_true, "y_score": y_score})
if hasattr(cal_fact_aucpr, "__len__"):
metrics = df.groupby("task", sort=True).apply(lambda g: all_metrics(
y_true=g.y_true.values,
y_score=g.y_score.values,
cal_fact_aucpr_task=cal_fact_aucpr[g['task'].values[0]]))
else:
metrics = df.groupby(
"task",
sort=True).apply(lambda g: all_metrics(y_true=g.y_true.values,
y_score=g.y_score.values,
cal_fact_aucpr_task=1.0))
metrics.reset_index(level=-1, drop=True, inplace=True)
return metrics.reindex(np.arange(num_tasks))
def compute_metrics(cols, y_true, y_score, num_tasks):
if len(cols) < 1:
return pd.DataFrame(
{
"roc_auc_score": np.nan,
"auc_pr": np.nan,
"avg_prec_score": np.nan,
"max_f1_score": np.nan,
"kappa": np.nan
},
index=np.arange(num_tasks))
df = pd.DataFrame({"task": cols, "y_true": y_true, "y_score": y_score})
metrics = df.groupby("task", sort=True).apply(lambda g: all_metrics(
y_true=g.y_true.values, y_score=g.y_score.values))
metrics.reset_index(level=-1, drop=True, inplace=True)
return metrics.reindex(np.arange(num_tasks))
def compute_metrics_regr(cols, y_true, y_score, num_tasks, y_censor=None):
"""Returns metrics for regression tasks."""
if len(cols) < 1:
return pd.DataFrame(
{
"rmse": np.nan,
"rmse_uncen": np.nan,
"rsquared": np.nan,
"corrcoef": np.nan,
},
index=np.arange(num_tasks))
df = pd.DataFrame({
"task": cols,
"y_true": y_true,
"y_score": y_score,
"y_censor": y_censor,
})
metrics = df.groupby("task", sort=True).apply(lambda g: all_metrics_regr(
y_true=g.y_true.values,
y_score=g.y_score.values,
y_censor=g.y_censor.values if y_censor is not None else None))
metrics.reset_index(level=-1, drop=True, inplace=True)
return metrics.reindex(np.arange(num_tasks))