You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by "WeichenXu123 (via GitHub)" <gi...@apache.org> on 2025/03/14 09:47:23 UTC
[PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
WeichenXu123 opened a new pull request, #50278:
URL: https://github.com/apache/spark/pull/50278
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines: https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR: https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g., '[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a faster review.
7. If you want to add a new configuration, please read the guideline first for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the guideline first in
'common/utils/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
Implement model size estimation.
2 new interfaces are added:
1. `Estimator.estimateModelSize`: This method is used to estimate the model size **before training**. This is an optional interface to be implemented. For linear classification & regression models the method is implemented. For other model such tree model, estimating model size before training is hard, so it is not implemented.
2. `Model.estimatedSize`: This method is used to estimate model size in **local process memory**. The default implementation is to use `SizeEstimator.estimate`, you can overwrite it to achieve more precise estimation.
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
3. If you fix a bug, you can clarify why it is a bug.
-->
For SparkConnect, we want to support model size control in spark connect server side. So this change is needed.
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as new features, bug fixes, or other behavior changes. Documentation-only updates are not considered user-facing changes.
If yes, please clarify the previous behavior and the change this PR proposes - provide the console output, description and/or an example to show the behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
No.
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some test cases that check the changes thoroughly including negative and positive cases if possible.
If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why it was difficult to add.
If benchmark tests were added, please run the benchmarks in GitHub Actions for the consistent environment, and the instructions could accord to: https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
No.
### Was this patch authored or co-authored using generative AI tooling?
<!--
If generative AI tooling has been used in the process of authoring this patch, please include the
phrase: 'Generated-by: ' followed by the name of the tool and its version.
If no, write 'No'.
Please refer to the [ASF Generative Tooling Guidance](https://www.apache.org/legal/generative-tooling.html) for details.
-->
No.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "WeichenXu123 (via GitHub)" <gi...@apache.org>.
WeichenXu123 commented on code in PR #50278:
URL: https://github.com/apache/spark/pull/50278#discussion_r2002620949
##########
mllib/src/main/scala/org/apache/spark/ml/Estimator.scala:
##########
@@ -81,4 +81,25 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
}
override def copy(extra: ParamMap): Estimator[M]
+
+ /**
+ * For ml connect only.
+ * Estimate an upper-bound size of the model to be fitted in bytes, based on the
+ * parameters and the dataset, e.g., using $(k) and numFeatures to estimate a
+ * k-means model size.
+ * 1, Only driver side memory usage is counted, distributed objects (like DataFrame,
+ * RDD, Graph, Summary) are ignored.
+ * 2, Lazy vals are not counted, e.g., an auxiliary object used in prediction.
+ * 3, If there is no enough information to get an accurate size, try to estimate the
+ * upper-bound size, e.g.
+ * - Given a LogisticRegression estimator, assume the coefficients are dense, even
+ * though the actual fitted model might be sparse (by L1 penalty).
+ * - Given a tree model, assume all underlying trees are complete binary trees, even
+ * though some branches might be pruned or truncated.
+ * 4, For some model such as tree model, estimating model size before training is hard,
+ * the `estimateModelSize` method is not supported.
+ */
+ private[spark] def estimateModelSize(dataset: Dataset[_]): Long = {
+ throw new UnsupportedOperationException
Review Comment:
yes.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "zhengruifeng (via GitHub)" <gi...@apache.org>.
zhengruifeng commented on code in PR #50278:
URL: https://github.com/apache/spark/pull/50278#discussion_r2002206038
##########
mllib/src/main/scala/org/apache/spark/ml/Estimator.scala:
##########
@@ -81,4 +81,25 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
}
override def copy(extra: ParamMap): Estimator[M]
+
+ /**
+ * For ml connect only.
+ * Estimate an upper-bound size of the model to be fitted in bytes, based on the
+ * parameters and the dataset, e.g., using $(k) and numFeatures to estimate a
+ * k-means model size.
+ * 1, Only driver side memory usage is counted, distributed objects (like DataFrame,
+ * RDD, Graph, Summary) are ignored.
+ * 2, Lazy vals are not counted, e.g., an auxiliary object used in prediction.
+ * 3, If there is no enough information to get an accurate size, try to estimate the
+ * upper-bound size, e.g.
+ * - Given a LogisticRegression estimator, assume the coefficients are dense, even
+ * though the actual fitted model might be sparse (by L1 penalty).
+ * - Given a tree model, assume all underlying trees are complete binary trees, even
+ * though some branches might be pruned or truncated.
+ * 4, For some model such as tree model, estimating model size before training is hard,
+ * the `estimateModelSize` method is not supported.
+ */
+ private[spark] def estimateModelSize(dataset: Dataset[_]): Long = {
+ throw new UnsupportedOperationException
Review Comment:
how to check whether an estimator implement this method in caller?
by try-catch?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "WeichenXu123 (via GitHub)" <gi...@apache.org>.
WeichenXu123 commented on code in PR #50278:
URL: https://github.com/apache/spark/pull/50278#discussion_r1995269037
##########
mllib/src/main/scala/org/apache/spark/ml/Estimator.scala:
##########
@@ -81,4 +81,25 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
}
override def copy(extra: ParamMap): Estimator[M]
+
+ /**
+ * For ml connect only.
+ * Estimate an upper-bound size of the model to be fitted in bytes, based on the
+ * parameters and the dataset, e.g., using $(k) and numFeatures to estimate a
+ * k-means model size.
+ * 1, Only driver side memory usage is counted, distributed objects (like DataFrame,
+ * RDD, Graph, Summary) are ignored.
+ * 2, Lazy vals are not counted, e.g., an auxiliary object used in prediction.
+ * 3, If there is no enough information to get an accurate size, try to estimate the
+ * upper-bound size, e.g.
+ * - Given a LogisticRegression estimator, assume the coefficients are dense, even
+ * though the actual fitted model might be sparse (by L1 penalty).
+ * - Given a tree model, assume all underlying trees are complete binary trees, even
+ * though some branches might be pruned or truncated.
+ * 4, For some model such as tree model, estimating model size before training is hard,
+ * the `estimateModelSize` method is not supported.
+ */
+ private[spark] def estimateModelSize(dataset: Dataset[_]): Long = {
+ throw new UnsupportedOperationException
Review Comment:
We don't need to implement this method for every model.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "WeichenXu123 (via GitHub)" <gi...@apache.org>.
WeichenXu123 closed pull request #50278: [SPARK-51340][ML][CONNECT] Model size estimation
URL: https://github.com/apache/spark/pull/50278
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "WeichenXu123 (via GitHub)" <gi...@apache.org>.
WeichenXu123 commented on PR #50278:
URL: https://github.com/apache/spark/pull/50278#issuecomment-2735618070
merged to master
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "WeichenXu123 (via GitHub)" <gi...@apache.org>.
WeichenXu123 commented on code in PR #50278:
URL: https://github.com/apache/spark/pull/50278#discussion_r1995257146
##########
mllib/src/main/scala/org/apache/spark/ml/Model.scala:
##########
@@ -43,4 +44,18 @@ abstract class Model[M <: Model[M]] extends Transformer {
def hasParent: Boolean = parent != null
override def copy(extra: ParamMap): M
+
+ /**
+ * For ml connect only.
+ * Estimate the size of this model in bytes.
+ * This is an approximation, the real size might be different.
+ * 1, Only driver side memory usage is counted, distributed objects (like DataFrame,
+ * RDD, Graph, Summary) are ignored.
+ * 2, Lazy vals are not counted, e.g., an auxiliary object used in prediction.
+ * 3, The default implementation uses `org.apache.spark.util.SizeEstimator.estimate`,
+ * some models override the default implementation to achieve more precise estimation.
+ * 4, For 3-rd extension, if external languages are used, it is recommended to override
+ * this method and return a proper size.
+ */
+ private[spark] def estimatedSize: Long = SizeEstimator.estimate(self)
Review Comment:
The default implementation should be ok in most cases. We don't need very accurate estimation, because the model size control is for restricting large model of Gigabytes order of magnitude. @zhengruifeng
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "zhengruifeng (via GitHub)" <gi...@apache.org>.
zhengruifeng commented on code in PR #50278:
URL: https://github.com/apache/spark/pull/50278#discussion_r2002206038
##########
mllib/src/main/scala/org/apache/spark/ml/Estimator.scala:
##########
@@ -81,4 +81,25 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
}
override def copy(extra: ParamMap): Estimator[M]
+
+ /**
+ * For ml connect only.
+ * Estimate an upper-bound size of the model to be fitted in bytes, based on the
+ * parameters and the dataset, e.g., using $(k) and numFeatures to estimate a
+ * k-means model size.
+ * 1, Only driver side memory usage is counted, distributed objects (like DataFrame,
+ * RDD, Graph, Summary) are ignored.
+ * 2, Lazy vals are not counted, e.g., an auxiliary object used in prediction.
+ * 3, If there is no enough information to get an accurate size, try to estimate the
+ * upper-bound size, e.g.
+ * - Given a LogisticRegression estimator, assume the coefficients are dense, even
+ * though the actual fitted model might be sparse (by L1 penalty).
+ * - Given a tree model, assume all underlying trees are complete binary trees, even
+ * though some branches might be pruned or truncated.
+ * 4, For some model such as tree model, estimating model size before training is hard,
+ * the `estimateModelSize` method is not supported.
+ */
+ private[spark] def estimateModelSize(dataset: Dataset[_]): Long = {
+ throw new UnsupportedOperationException
Review Comment:
in caller, how to check whether an estimator implement this method?
by try-catch?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] [SPARK-51340][ML][CONNECT] Model size estimation [spark]
Posted by "WeichenXu123 (via GitHub)" <gi...@apache.org>.
WeichenXu123 commented on code in PR #50278:
URL: https://github.com/apache/spark/pull/50278#discussion_r1995269037
##########
mllib/src/main/scala/org/apache/spark/ml/Estimator.scala:
##########
@@ -81,4 +81,25 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
}
override def copy(extra: ParamMap): Estimator[M]
+
+ /**
+ * For ml connect only.
+ * Estimate an upper-bound size of the model to be fitted in bytes, based on the
+ * parameters and the dataset, e.g., using $(k) and numFeatures to estimate a
+ * k-means model size.
+ * 1, Only driver side memory usage is counted, distributed objects (like DataFrame,
+ * RDD, Graph, Summary) are ignored.
+ * 2, Lazy vals are not counted, e.g., an auxiliary object used in prediction.
+ * 3, If there is no enough information to get an accurate size, try to estimate the
+ * upper-bound size, e.g.
+ * - Given a LogisticRegression estimator, assume the coefficients are dense, even
+ * though the actual fitted model might be sparse (by L1 penalty).
+ * - Given a tree model, assume all underlying trees are complete binary trees, even
+ * though some branches might be pruned or truncated.
+ * 4, For some model such as tree model, estimating model size before training is hard,
+ * the `estimateModelSize` method is not supported.
+ */
+ private[spark] def estimateModelSize(dataset: Dataset[_]): Long = {
+ throw new UnsupportedOperationException
Review Comment:
We don't need to implement this method for every model, it is nice to have, but if no implementation, we can still compute model size from the trained model.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org