def update_state(self, y_true, y_pred, sample_weight=None): y_pred = tf.round(y_pred) y_pred = tf.cast(y_pred, tf.bool) y_true = tf.cast(y_true, tf.bool) values_tn = tf.logical_and(tf.equal(y_true, False), tf.equal(y_pred, False)) values_tn = tf.cast(values_tn, self.dtype) values_fp = tf.logical_and(tf.equal(y_true, False), tf.equal(y_pred, True)) values_fp = tf.cast(values_fp, self.dtype) values_fn = tf.logical_and(tf.equal(y_true, True), tf.equal(y_pred, False)) values_fn = tf.cast(values_fn, self.dtype) values_tp = tf.logical_and(tf.equal(y_true, True), tf.equal(y_pred, True)) values_tp = tf.cast(values_tp, self.dtype) if sample_weight is not None: sample_weight = tf.cast(sample_weight, self.dtype) sample_weight_tn = tf.broadcast_weights(sample_weight, values_tn) sample_weight_fp = tf.broadcast_weights(sample_weight, values_fp) sample_weight_fn = tf.broadcast_weights(sample_weight, values_fn) sample_weight_tp = tf.broadcast_weights(sample_weight, values_tp) values_tn = tf.multiply(values_tn, sample_weight_tn) values_fp = tf.multiply(values_fp, sample_weight_fp) values_fn = tf.multiply(values_fn, sample_weight_fn) values_tp = tf.multiply(values_tp, sample_weight_tp) self.tn.assign_add(tf.reduce_sum(values_tn)) self.fp.assign_add(tf.reduce_sum(values_fp)) self.fn.assign_add(tf.reduce_sum(values_fn)) self.tp.assign_add(tf.reduce_sum(values_tp))
def correlation_thresh(y_true, y_pred, thresh=2.58, sample_weight=None):
y_true = tf.cast(y_true, tf.float32)
y_pred = tf.cast(y_pred, tf.float32)
#### THRESHOLD DATA ####
data_intersect = tf.cast(tf.math.greater(y_true, thresh),
dtype=tf.float32) * y_pred
mask_intersect = tf.cast(data_intersect, dtype=tf.bool)
y_true = tf.boolean_mask(y_true, mask_intersect)
y_pred = tf.boolean_mask(y_pred, mask_intersect)
mean_ytrue = tf.reduce_mean(y_true, keepdims=True)
mean_ypred = tf.reduce_mean(y_pred, keepdims=True)
demean_ytrue = y_true - mean_ytrue
demean_ypred = y_pred - mean_ypred
if sample_weight is not None:
sample_weight = tf.broadcast_weights(sample_weight, y_true)
std_y = tf.sqrt(
tf.reduce_sum(sample_weight * tf.square(demean_ytrue)) *
tf.reduce_sum(sample_weight * tf.square(demean_ypred)))
correlation = tf.reduce_sum(
sample_weight * demean_ytrue * demean_ypred) / std_y
else:
std_y = tf.sqrt(
tf.reduce_sum(tf.square(demean_ytrue)) *
tf.reduce_sum(tf.square(demean_ypred)))
correlation = tf.reduce_sum(demean_ytrue * demean_ypred) / std_y
#tf.print("correlation:", correlation,output_stream=sys.stdout)
return tf.maximum(tf.minimum(correlation, 1.0), -1.0)
def update_state(self, y_true, y_pred, sample_weight=None):
'''
Arguments:
y_true The actual y. Passed by default to Metric classes.
y_pred The predicted y. Passed by default to Metric classes.
'''
# Compute the number of negatives.
y_true = tf.cast(y_true, tf.bool)
actual_positives = tf.reduce_sum(
tf.cast(tf.equal(y_true, True), self.dtype))
self.actual_positives.assign_add(actual_positives)
# Compute the number of false positives.
y_pred = tf.greater_equal(
y_pred, 0.5
) # Using default threshold of 0.5 to call a prediction as positive labeled.
predicted_positives = tf.reduce_sum(
tf.cast(tf.equal(y_pred, True), self.dtype))
self.predicted_positives.assign_add(predicted_positives)
true_positive_values = tf.logical_and(tf.equal(y_true, True),
tf.equal(y_pred, True))
true_positive_values = tf.cast(true_positive_values, self.dtype)
if sample_weight is not None:
sample_weight = tf.cast(sample_weight, self.dtype)
sample_weight = tf.broadcast_weights(sample_weight, values)
values = tf.multiply(true_positive_values, sample_weight)
true_positives = tf.reduce_sum(true_positive_values)
self.true_positives.assign_add(true_positives)
def correlation(y_true, y_pred, sample_weight=None):
#### GET RID OF ZEROS ####
y_true = tf.cast(y_true, tf.float32)
y_pred = tf.cast(y_pred, tf.float32)
data_intersect = y_true * y_pred
mask_intersect = tf.cast(data_intersect, dtype=tf.bool)
y_true = tf.boolean_mask(y_true, mask_intersect)
y_pred = tf.boolean_mask(y_pred, mask_intersect)
mean_ytrue = tf.reduce_mean(y_true, keepdims=True)
mean_ypred = tf.reduce_mean(y_pred, keepdims=True)
demean_ytrue = y_true - mean_ytrue
demean_ypred = y_pred - mean_ypred
if sample_weight is not None:
sample_weight = tf.broadcast_weights(sample_weight, y_true)
std_y = tf.sqrt(
tf.reduce_sum(sample_weight * tf.square(demean_ytrue)) *
tf.reduce_sum(sample_weight * tf.square(demean_ypred)))
correlation = tf.reduce_sum(
sample_weight * demean_ytrue * demean_ypred) / std_y
else:
std_y = tf.sqrt(
tf.reduce_sum(tf.square(demean_ytrue)) *
tf.reduce_sum(tf.square(demean_ypred)))
correlation = tf.reduce_sum(demean_ytrue * demean_ypred) / std_y
return tf.maximum(tf.minimum(correlation, 1.0), -1.0)
def update_state(self, y_true, y_pred, sample_weight = None):
y_true = tf.cast(y_true[:, 0], tf.bool)
y_pred = tf.math.greater_equal(y_pred[:, 0], self.thresholds)
values = tf.logical_and(tf.equal(y_true, False), tf.equal(y_pred, True))
values = tf.cast(values, self.dtype)
if sample_weight is not None:
sample_weight = tf.cast(sample_weight, self.dtype)
sample_weight = tf.broadcast_weights(sample_weight, values)
values = tf.math.multiply(values, sample_weight)
self.false_positives.assign_add(tf.reduce_sum(values))
def update_state(self, y_true, y_pred, sample_weight=None):
"""Update confusion matrix after epoch"""
y_true = tf.argmax(y_true, axis=1)
y_true = tf.cast(y_true, tf.int32)
y_pred = tf.argmax(y_pred, axis=1)
y_pred = tf.cast(y_pred, tf.int32)
conf_mat = tf.math.confusion_matrix(y_true,
y_pred,
num_classes=self.num_classes,
dtype=tf.float32)
if sample_weight:
sample_weight = tf.cast(sample_weight, self.dtype)
sample_weight = tf.broadcast_weights(sample_weight, conf_mat)
conf_mat = tf.multiply(conf_mat, sample_weight)
self.conf_mat.assign_add(conf_mat)
def calculate_correlation(y_true, y_pred, sample_weight=None):
assert len(y_true.shape) == 5
mean_ytrue = tf.reduce_mean(y_true, keepdims=True, axis=[1, 2, 3, 4])
mean_ypred = tf.reduce_mean(y_pred, keepdims=True, axis=[1, 2, 3, 4])
demean_ytrue = y_true - mean_ytrue
demean_ypred = y_pred - mean_ypred
if sample_weight is not None:
sample_weight = tf.broadcast_weights(sample_weight, y_true)
std_y = tf.sqrt(
tf.reduce_sum(sample_weight * tf.square(demean_ytrue)) *
tf.reduce_sum(sample_weight * tf.square(demean_ypred)))
correlation = tf.reduce_sum(
sample_weight * demean_ytrue * demean_ypred) / std_y
else:
std_y = tf.sqrt(
tf.reduce_sum(tf.square(demean_ytrue)) *
tf.reduce_sum(tf.square(demean_ypred)))
correlation = tf.reduce_sum(demean_ytrue * demean_ypred) / std_y
return tf.maximum(tf.minimum(correlation, 1.0), -1.0)
def correlation_gm(y_true, y_pred, sample_weight=None):
sz = K.ndim(y_true)
gm = nibabel.load(
'/data/Templates/Yeo2011_17Networks_2mm_LiberalMask_64.nii.gz'
).get_fdata()
if K.eval(sz) == 5:
gm = np.expand_dims(gm, axis=[0, -1])
gm = tf.cast(gm, tf.bool)
#### GM Mask ####
y_true = tf.boolean_mask(y_true, gm)
y_pred = tf.boolean_mask(y_pred, gm)
mean_ytrue = tf.reduce_mean(y_true, keepdims=True)
mean_ypred = tf.reduce_mean(y_pred, keepdims=True)
demean_ytrue = y_true - mean_ytrue
demean_ypred = y_pred - mean_ypred
if sample_weight is not None:
sample_weight = tf.broadcast_weights(sample_weight, y_true)
std_y = tf.sqrt(
tf.reduce_sum(sample_weight * tf.square(demean_ytrue)) *
tf.reduce_sum(sample_weight * tf.square(demean_ypred)))
correlation = tf.reduce_sum(
sample_weight * demean_ytrue * demean_ypred) / std_y
else:
std_y = tf.sqrt(
tf.reduce_sum(tf.square(demean_ytrue)) *
tf.reduce_sum(tf.square(demean_ypred)))
correlation = tf.reduce_sum(demean_ytrue * demean_ypred) / std_y
return tf.maximum(tf.minimum(correlation, 1.0), -1.0)
