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):
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
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))
def print_metrics_to_latex(metrics, filename):
r"""Print metrics to latex and format them as \\bscellA{}."""
metrics = metrics.copy()
def cell_format(value, char):
return '\\bscell%s[%.3f]{%3.0f}' % (char, value, value * 100)
def float_format_short(value):
return '%.3f' % (value, )
def float_format_long(value):
return '%.4f' % (value, )
def float_formatA(value):
return cell_format(value, 'A')
def float_formatB(value):
return cell_format(value, 'B')
# use formatB for all 'ROC' columns
formatters = {}
for value in metrics.columns.values:
if value[1] == 'PR':
formatters[value] = float_formatA
elif value[1] == 'ROC':
formatters[value] = float_formatB
elif value[1] == 'ACC':
formatters[value] = float_format_long
else:
formatters[value] = None
# workaround because the index is not properly formatted in Latex
metrics = metrics.reset_index()
metrics.to_latex(filename,
float_format=float_format_short,
formatters=formatters,
escape=False,
index=False)