add introduction of PNP Loss

docs/extend/losses.md

@@ -93,7 +93,7 @@ There are also a few functions in ```self.distance``` that provide some of this
 
 ## Using ```indices_tuple```
 
-This is an optional argument passed in from the outside. (See the [overview](../../#using-losses-and-miners-in-your-training-loop) for an example.) It currently has 3 possible forms: 
+This is an optional argument passed in from the outside. (See the [overview](../index.md#using-losses-and-miners-in-your-training-loop) for an example.) It currently has 3 possible forms: 
 
  - ```None```
  - A tuple of size 4, representing the indices of mined pairs (anchors, positives, anchors, negatives)

docs/index.md

@@ -13,7 +13,7 @@ This library contains 9 modules, each of which can be used independently within
 ## How loss functions work
 
 ### Using losses and miners in your training loop
-Let’s initialize a plain [TripletMarginLoss](losses/#tripletmarginloss):
+Let’s initialize a plain [TripletMarginLoss](losses.md#tripletmarginloss):
 ```python
 from pytorch_metric_learning import losses
 loss_func = losses.TripletMarginLoss()
@@ -95,8 +95,8 @@ If you're interested in [MoCo](https://arxiv.org/pdf/1911.05722.pdf)-style self-
 
 ## Highlights of the rest of the library
 
-- For a convenient way to train your model, take a look at the [trainers](trainers).
-- Want to test your model's accuracy on a dataset? Try the [testers](testers/).
+- For a convenient way to train your model, take a look at the [trainers](trainers.md).
+- Want to test your model's accuracy on a dataset? Try the [testers](testers.md).
 - To compute the accuracy of an embedding space directly, use [AccuracyCalculator](accuracy_calculation.md).
 
 If you're short of time and want a complete train/test workflow, check out the [example Google Colab notebooks](https://github.com/KevinMusgrave/pytorch-metric-learning/tree/master/examples).

docs/losses.md

@@ -889,6 +889,30 @@ loss = loss_fn(embeddings, labels)
 ```python
 losses.PNPLoss(b=2, alpha=1, anneal=0.01, variant="O", **kwargs)
 ```
+**Equation**:
+
+![PNP_loss_equation](imgs/PNP_loss_equation.png){: style="height:300px"}
+
+**Parameters**:
+
+* **b**: The boundary of PNP-Ib (see equation 9 above). The paper uses 2.
+* **alpha**: The power of PNP-Dq (see equation 13 above). The paper uses 8.
+* **anneal**: The temperature of the sigmoid function. (The sigmoid function is `R` in the equations above.) The paper uses 0.01.
+* **variant**: The name of the variant. The options are {"Ds", "Dq", "Iu", "Ib", "O"}. The paper uses "Dq".
+
+**Default distance**:
+
+- [```CosineSimilarity()```](distances.md#cosinesimilarity)
+    - This is the only compatible distance.
+
+**Default reducer**: 
+
+ - [MeanReducer](reducers.md#meanreducer)
+
+**Reducer input**:
+
+* **loss**: The loss per element that has at least 1 positive in the batch. Reduction type is ```"element"```.
+
 
 
 ## ProxyAnchorLoss