def Accuracy(y_true, y_pred, multioutput=None):
"""
Calculate the accuracy score (Accuracy).
.. math::
\\text{Accuracy} = \\frac{1}{n}\\sum_{t=1}^n 1(y_t=\\hat{y_t})
:param y_true: Array-like of shape = (n_samples, \*).
Ground truth (correct) target values.
:param y_pred: Array-like of shape = (n_samples, \*).
Estimated target values.
:return: Float or ndarray of floats.
A non-negative floating point value (the best value is 1.0), or an
array of floating point values, one for each individual target.
"""
from sklearn.metrics._classification import accuracy_score
y_true = np.squeeze(y_true)
y_pred = np.squeeze(y_pred)
if np.any(y_pred != y_pred.astype(int)):
# y_pred is probability
if y_pred.ndim == 1:
y_pred = np.where(y_pred > 0.5, 1, 0)
else:
y_pred = np.argmax(y_pred, axis=1)
return accuracy_score(y_true, y_pred)
def Accuracy(y_true, y_pred, multioutput=None):
from sklearn.metrics._classification import accuracy_score
y_true = np.squeeze(y_true)
y_pred = np.squeeze(y_pred)
if np.any(y_pred != y_pred.astype(int)):
# y_pred is probability
if y_pred.ndim == 1:
y_pred = np.where(y_pred > 0.5, 1, 0)
else:
y_pred = np.argmax(y_pred, axis=1)
return accuracy_score(y_true, y_pred)
def experiment_transductive(name='E(1)', exp_index=0, preprocessing='none'):
"""
performs transductive experiment for frame/target pair
parameters:
name: image name
exp_index: no. experiment (e.g. 0-9)
preprocessing: preprocessing mode
returns:
True/False
"""
start = time.time()
outfile = 'results/{}_{}_{}.npz'.format(name, exp_index, preprocessing)
data, train_gt, test_gt, _ = load_ds(name, exp_index)
#preprocessing
X = data.reshape(-1, data.shape[2])
X = preprocess(X, preprocessing=preprocessing)
data = X.reshape(data.shape)
X_train = data[train_gt > 0]
y_train = train_gt[train_gt > 0]
time_prepare = time.time() - start
start = time.time()
svm = TwoStageSVM()
svm.fit(X_train, y_train)
time_train = time.time() - start
X_test = data[test_gt > 0]
y_test = test_gt[test_gt > 0]
start = time.time()
y_pred = svm.predict(X_test)
time_test = time.time() - start
y_pred_all = svm.predict(X)
acc = accuracy_score(y_test, y_pred)
np.savez_compressed(outfile,
y_test=y_test,
y_pred=y_pred,
sh=data.shape,
best_params=[svm.params_['C'], svm.params_['gamma']],
acc=[acc],
cls_score=[svm.score_],
y_pred_all=y_pred_all,
time_train=time_train,
time_test=time_test,
time_prepare=time_prepare)
print(outfile, acc, time_train, time_test)
def Accuracy(y_true, y_pred, multioutput=None):
from sklearn.metrics._classification import accuracy_score
return accuracy_score(y_true, y_pred)
def experiment_inductive(name_train='hyperblood_frame_day_1_afternoon',
name_test='hyperblood_comparison_day_1',
exp_index=0,
preprocessing='none'):
"""
performs inductive experiment for frame/target pair
parameters:
name_train: image name (for training)
name_test: image name (for testing)
exp_index: no. experiment (e.g. 0-9)
preprocessing: preprocessing mode
"""
print(name_train, name_test)
start = time.time()
outfile = 'results_i/{}_{}_{}_{}.npz'.format(name_train, name_test,
exp_index, preprocessing)
data_train, train_gt, _, _ = load_ds(name_train, exp_index)
data_test, _, test_gt, _ = load_ds(name_test, exp_index)
_, anno_test, _, _, _ = blood_loader(PATH_DATA, name_test)
test_gt = np.asarray(test_gt, dtype=np.int32)
anno_test = np.asarray(anno_test, dtype=np.int32)
assert np.sum(np.unique(anno_test) - np.unique(test_gt)) == 0
#preprocessing
X = data_train.reshape(-1, data_train.shape[2])
X = preprocess(X, preprocessing=preprocessing)
data_train = X.reshape(data_train.shape)
X = data_test.reshape(-1, data_test.shape[2])
X = preprocess(X, preprocessing=preprocessing)
data_test = X.reshape(data_test.shape)
X_train = data_train[train_gt > 0]
y_train = train_gt[train_gt > 0]
time_prepare = time.time() - start
start = time.time()
svm = TwoStageSVM()
svm.fit(X_train, y_train)
time_train = time.time() - start
#second version of results
X_test_gt = data_test.copy()[anno_test > 0]
y_test_gt = anno_test.copy()[anno_test > 0]
X_test = data_test[test_gt > 0]
y_test = test_gt[test_gt > 0]
y_pred = svm.predict(X_test)
y_pred_all = svm.predict(X)
start = time.time()
y_pred_gt = svm.predict(X_test_gt)
time_test = time.time() - start
acc = accuracy_score(y_test, y_pred)
acc_gt = accuracy_score(y_test_gt, y_pred_gt)
np.savez_compressed(outfile,
y_test=y_test,
y_pred=y_pred,
sh_train=data_train.shape,
sh_test=data_test.shape,
best_params=[svm.params_['C'], svm.params_['gamma']],
acc=[acc],
cls_score=[svm.score_],
y_pred_all=y_pred_all,
y_test_gt=y_test_gt,
y_pred_gt=y_pred_gt,
acc_gt=[acc_gt],
time_train=time_train,
time_test=time_test,
time_prepare=time_prepare)
print(outfile, acc, acc_gt, time_train, time_test)
