def _mostrar_resultados(y_test, pred_y):
conf_matrix = _confusion_matrix(y_test, pred_y)
_plt.figure(figsize=(10, 10))
_sns.heatmap(conf_matrix,
xticklabels=LABELS,
yticklabels=LABELS,
annot=True,
fmt="d")
_plt.title("Confusion matrix")
_plt.ylabel('True class')
_plt.xlabel('Predicted class')
print(_classification_report(y_test, pred_y))
def _getConfusionMatrix(self, ):
"""
Calculates confusion matrix from predictions and truth labels.
Parameters
==========
None.
Returns
=======
numpy 2-d array.
"""
csv_buffer = StringIO()
labels = set(self.truths) | set(self.preds)
cm = _confusion_matrix(self.truths, self.preds, labels=list(labels))
pd.DataFrame(cm, columns=labels, index=labels).to_csv(csv_buffer)
return csv_buffer.getvalue()
def confusion_matrix(
orig: Union[pd.Series, np.ndarray, Sequence],
new: Union[pd.Series, np.ndarray, Sequence],
data: Optional[pd.DataFrame] = None,
*,
normalize: bool = True,
) -> pd.DataFrame:
"""\
Given an original and new set of labels, create a labelled confusion matrix.
Parameters `orig` and `new` can either be entries in data or categorical arrays
of the same size.
Params
------
orig
Original labels.
new
New labels.
data
Optional dataframe to fill entries from.
normalize
Should the confusion matrix be normalized?
Examples
--------
.. plot::
import scanpy as sc; import seaborn as sns
pbmc = sc.datasets.pbmc68k_reduced()
cmtx = sc.metrics.confusion_matrix("bulk_labels", "louvain", pbmc.obs)
sns.heatmap(cmtx)
"""
from sklearn.metrics import confusion_matrix as _confusion_matrix
if data is not None:
if isinstance(orig, str):
orig = data[orig]
if isinstance(new, str):
new = data[new]
# Coercing so I don't have to deal with it later
orig, new = pd.Series(orig), pd.Series(new)
assert len(orig) == len(new)
unique_labels = pd.unique(np.concatenate((orig.values, new.values)))
# Compute
mtx = _confusion_matrix(orig, new, labels=unique_labels)
if normalize:
sums = mtx.sum(axis=1)[:, np.newaxis]
mtx = np.divide(mtx, sums, where=sums != 0)
# Label
orig_name = "Original labels" if orig.name is None else orig.name
new_name = "New Labels" if new.name is None else new.name
df = pd.DataFrame(
mtx,
index=pd.Index(unique_labels, name=orig_name),
columns=pd.Index(unique_labels, name=new_name),
)
# Filter
if is_categorical(orig):
orig_idx = pd.Series(orig).cat.categories
else:
orig_idx = natsorted(pd.unique(orig))
if is_categorical(new):
new_idx = pd.Series(new).cat.categories
else:
new_idx = natsorted(pd.unique(new))
df = df.loc[np.array(orig_idx), np.array(new_idx)]
return df
def confusion_matrix(y, y_pred):
cm = _confusion_matrix(y, y_pred)
cm = (cm.T / cm.sum(axis=1)).T
return cm
def confusion_matrix(Y_true, Y_pred, labels=None):
if labels is None:
labels = np.unique(np.append(Y_true, Y_pred))
return pd.DataFrame(_confusion_matrix(Y_true, Y_pred, labels=labels),
index=[f'actual_{label}' for label in labels],
columns=[f'predicted_{label}' for label in labels])