Internal change

PiperOrigin-RevId: 410043098

tensorflow_ranking/python/keras/losses.py

@@ -668,9 +668,9 @@ class ListMLELoss(_ListwiseLoss):
   \mathcal{L}(\{y\}, \{s\}) = - \log(P(\pi_y | s))
   $$
 
-  where $$P(\pi_y | s)$$ is the Plackett-Luce probability of a permutation
-  $$\pi_y$$ conditioned on scores $$s$$. Here $$\pi_y$$ represents a permutation
-  of items ordered by the relevance labels $$y$$ where ties are broken randomly.
+  where $P(\pi_y | s)$ is the Plackett-Luce probability of a permutation
+  $\pi_y$ conditioned on scores $s$. Here $\pi_y$ represents a permutation
+  of items ordered by the relevance labels $y$ where ties are broken randomly.
 
   References:
     - [Listwise approach to learning to rank: theory and algorithm, Xia et al,
@@ -963,7 +963,7 @@ class ClickEMLoss(_RankingLoss):
 
   Implementation of click EM loss ([Wang et al, 2018][wang2018]). This loss
   assumes that a click is generated by a factorized model
-  $$P(\text{examination}) \cdot P(\text{relevance})$$, which are latent
+  $P(\text{examination}) \cdot P(\text{relevance})$, which are latent
   variables determined by `exam_logits` and `rel_logits` respectively.
 
   NOTE: This loss should be called with a `logits` tensor of shape
@@ -1043,7 +1043,7 @@ class SigmoidCrossEntropyLoss(_RankingLoss):
   ```
 
   NOTE: This loss does not support graded relevance labels and should only be
-  used with binary relevance labels ($$y \in [0, 1]$$).
+  used with binary relevance labels ($y \in [0, 1]$).
 
   Standalone usage:
 

tensorflow_ranking/python/keras/metrics.py

@@ -237,8 +237,8 @@ class MRRMetric(_RankingMetric):
   \text{MRR}(\{y\}, \{s\}) = \max_i \frac{\bar{y}_i}{\text{rank}(s_i)}
   $$
 
-  where $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores
-  $$s$$ with ties broken randomly and $$\bar{y_i}$$ are truncated labels:
+  where $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores
+  $s$ with ties broken randomly and $\bar{y_i}$ are truncated labels:
 
   $$
   \bar{y}_i = \begin{cases}
@@ -300,8 +300,8 @@ class ARPMetric(_RankingMetric):
   \frac{1}{\sum_i y_i} \sum_i y_i \cdot \text{rank}(s_i)
   $$
 
-  where $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores
-  $$s$$ with ties broken randomly.
+  where $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores
+  $s$ with ties broken randomly.
   """
 
   def __init__(self, name=None, dtype=None, ragged=False, **kwargs):
@@ -353,21 +353,21 @@ class PrecisionMetric(_RankingMetric):
 
   where:
 
-  * $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores $$s$$
+  * $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores $s$
     with ties broken randomly
-  * $$I[]$$ is the indicator function:\
-    $$I[\text{cond}] = \begin{cases}
+  * $I[]$ is the indicator function:\
+    $I[\text{cond}] = \begin{cases}
     1 & \text{if cond is true}\\
     0 & \text{else}\end{cases}
-    $$
-  * $$\bar{y}_i$$ are the truncated labels:\
-    $$
+    $
+  * $\bar{y}_i$ are the truncated labels:\
+    $
     \bar{y}_i = \begin{cases}
     1 & \text{if }y_i \geq 1 \\
     0 & \text{else}
     \end{cases}
-    $$
-  * $$k = |y|$$ if $$k$$ is not provided
+    $
+  * $k = |y|$ if $k$ is not provided
   """
 
   def __init__(self, name=None, topn=None, dtype=None, ragged=False, **kwargs):
@@ -429,21 +429,21 @@ class RecallMetric(_RankingMetric):
 
   where:
 
-  * $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores $$s$$
+  * $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores $s$
     with ties broken randomly
-  * $$I[]$$ is the indicator function:\
-    $$I[\text{cond}] = \begin{cases}
+  * $I[]$ is the indicator function:\
+    $I[\text{cond}] = \begin{cases}
     1 & \text{if cond is true}\\
     0 & \text{else}\end{cases}
-    $$
-  * $$\bar{y}_i$$ are the truncated labels:\
-    $$
+    $
+  * $\bar{y}_i$ are the truncated labels:\
+    $
     \bar{y}_i = \begin{cases}
     1 & \text{if }y_i \geq 1 \\
     0 & \text{else}
     \end{cases}
-    $$
-  * $$k = |y|$$ if $$k$$ is not provided
+    $
+  * $k = |y|$ if $k$ is not provided
   """
 
   def __init__(self, name=None, topn=None, dtype=None, ragged=False, **kwargs):
@@ -512,8 +512,8 @@ class PrecisionIAMetric(_RankingMetric):
   \sum_t \sum_i I[\text{rank}(s_i) \leq k] y_{i,t}
   $$
 
-  where $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores
-  $$s$$ with ties broken randomly.
+  where $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores
+  $s$ with ties broken randomly.
 
   References:
 
@@ -603,22 +603,22 @@ class MeanAveragePrecisionMetric(_RankingMetric):
 
   where:
 
-  * $$P@k(y, s)$$ is the Precision at rank $$k$$. See
+  * $P@k(y, s)$ is the Precision at rank $k$. See
     `tfr.keras.metrics.PrecisionMetric`.
-  * $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores $$s$$
+  * $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores $s$
     with ties broken randomly
-  * $$I[]$$ is the indicator function:\
-    $$I[\text{cond}] = \begin{cases}
+  * $I[]$ is the indicator function:\
+    $I[\text{cond}] = \begin{cases}
     1 & \text{if cond is true}\\
     0 & \text{else}\end{cases}
-    $$
-  * $$\bar{y}_i$$ are the truncated labels:\
-    $$
+    $
+  * $\bar{y}_i$ are the truncated labels:\
+    $
     \bar{y}_i = \begin{cases}
     1 & \text{if }y_i \geq 1 \\
     0 & \text{else}
     \end{cases}
-    $$
+    $
   """
 
   def __init__(self, name=None, topn=None, dtype=None, ragged=False, **kwargs):
@@ -685,8 +685,8 @@ class NDCGMetric(_RankingMetric):
   \sum_i \text{gain}(y_i) \cdot \text{rank_discount}(\text{rank}(s_i))
   $$
 
-  where $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores
-  $$s$$ with ties broken randomly.
+  where $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores
+  $s$ with ties broken randomly.
 
   References:
 
@@ -771,8 +771,8 @@ class DCGMetric(_RankingMetric):
   \sum_i \text{gain}(y_i) \cdot \text{rank_discount}(\text{rank}(s_i))
   $$
 
-  where $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores
-  $$s$$ with ties broken randomly.
+  where $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores
+  $s$ with ties broken randomly.
 
   References:
 
@@ -868,8 +868,8 @@ class AlphaDCGMetric(_RankingMetric):
   y_{i, t} (1 - \alpha)^{\sum_j I[\text{rank}(s_j) < \text{rank}(s_i)] y_{j, t}}
   $$
 
-  where $$\text{rank}(s_i)$$ is the rank of item $$i$$ after sorting by scores
-  $$s$$ with ties broken randomly and $$I[]$$ is the indicator function:
+  where $\text{rank}(s_i)$ is the rank of item $i$ after sorting by scores
+  $s$ with ties broken randomly and $I[]$ is the indicator function:
 
   $$
   I[\text{cond}] = \begin{cases}
@@ -981,7 +981,7 @@ class OPAMetric(_RankingMetric):
   \frac{\sum_i \sum_j I[s_i > s_j] I[y_i > y_j]}{\sum_i \sum_j I[y_i > y_j]}
   $$
 
-  where $$I[]$$ is the indicator function:
+  where $I[]$ is the indicator function:
 
   $$
   I[\text{cond}] = \begin{cases}