Keras 損失函式與評價指標詳解

2020.12.03更新：pytorch不香麼…

regression losses

2、metrics

變數定義：

損失函式的分類：

probabilistic losses

binarycrossentropy class

#  輸入示例

#  y_true = [[0., 1.], [0., 0.]]
#  y_pred = [[0.6, 0.4], [0.4, 0.6]]
tf.keras.losses.binarycrossentropy(
from_logits=
false
, label_smoothing=
0, reduction=
"auto"
, name=
"binary_crossentropy"
)

# 輸入示例

# y_true = [[0, 1, 0], [0, 0, 1]]
# y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]]
tf.keras.losses.categoricalcrossentropy(
from_logits=
false
,    label_smoothing=0,
reduction=
"auto"
,    name=
"categorical_crossentropy"
,)

# y_true = [1, 2]

# y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]]
tf.keras.losses.sparsecategoricalcrossentropy(
from_logits=
false
, reduction=
"auto"
, name=
"sparse_categorical_crossentropy"
)

泊松損失

參考：泊松回歸詳細介紹

定義：

loss = y_pred - y_true * log(y_pred)

kl散度：

定義：loss = y_true * log(y_true / y_pred)

參考：regression losses

meansquarederror class

均方差

loss = square(y_true - y_pred)

絕對值均差

loss =

abs(y_true - y_pred)

絕對值相對差

loss =

100*
abs(y_true - y_pred)
/ y_true

詞窮了，咋翻譯？

loss = square(log(y_true +1.

)- log(y_pred +1.
))

余弦相似度

loss =

-sum
(l2_norm(y_true)
* l2_norm(y_pred)
)

去掉離群點，減輕離群點對loss的影響

loss =

0.5* x^2if
|x|<= d
loss =
0.5* d^
2+ d *
(|x|
- d)
if|x|
> d

logcosh = log(

(exp(x)
+ exp(
-x))/2
), where x is the error y_pred - y_true.

classification metrics based on true/false positives & negatives

auc class (area under the curve)

tf.keras.metrics.auc(

num_thresholds=
200,
curve=
"roc"
,    summation_method=
"interpolation"
,    name=
none
,    dtype=
none
,    thresholds=
none
,    multi_label=
false
,    label_weights=
none
,)

recall class

precisionatrecall class

計算當 recall is >= specified value 時的 precision

truepositives class，truenegatives class， falsepositives class， falsenegatives class

sensitivityatspecificity class

specificityatsensitivity class

Keras損失函式 評價標準

計算真實標籤和 標籤之間的交叉熵損失 當只有兩個標籤類 假設為0和1 時，使用這個交叉熵損失。對於每個示例，每個 都應該有乙個浮點值 tf.keras.losses.binary crossentropy y true tf.cast label,dtype tf.float32 y pred tf...

損失函式與評價函式

4.1 學習目標 掌握常見的評價函式和損失函式dice iou bce focal loss lov sz softmax 掌握評價 損失函式的實踐 4.2 tp tn fp fn 在講解語義分割中常用的評價函式和損失函式之前，先補充一 tp 真正例 true positive tn 真反例 tru...

Keras加權損失函式

ref keras提供的損失函式binary crossentropy和categorical crossentropy沒有加權，如果想實現樣本的不同權重功能有一種策略是對損失函式加權處理。二分加權交叉熵損失 class weightedbinarycrossentropy object def i...