Github user mengxr commented on a diff in the pull request:
https://github.com/apache/spark/pull/9229#discussion_r56207844
--- Diff: mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala ---
@@ -32,20 +32,43 @@ import org.apache.spark.util.random.XORShiftRandom
*
*/
private[ann] trait Layer extends Serializable {
+
+ /**
+ * Number of weights that is used to allocate memory for the weights
vector
+ */
+ val weightSize: Int
+
+ /**
+ * Returns the output size given the input size (not counting the stack
size).
+ * Output size is used to allocate memory for the output.
+ * @param inputSize input size
+ * @return output size
+ */
+ def outputSize(inputSize: Int): Int
+
+ /**
+ * If true, the memory is not allocated for the output of this layer.
+ * The memory allocated to the previous layer is used to write the
output of this layer.
+ * Developer can set this to true if computing delta of a previous layer
+ * does not involve its output, so the current layer can write there.
+ * This also mean that both layers have the same number of outputs.
+ */
+ val inPlace: Boolean
+
/**
- * Returns the instance of the layer based on weights provided
+ * Returns the instance of the layer based on weights provided.
+ * Size of weights must be equal to weightSize
* @param weights vector with layer weights
- * @param position position of weights in the vector
* @return the layer model
*/
- def getInstance(weights: Vector, position: Int): LayerModel
-
+ def model(weights: BDV[Double]): LayerModel
--- End diff --
* `createModel` and rename `weights` to `initialWeights`
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
