Currently I am studying the usage of Apache Spark 3.0 with Rapids GPU Acceleration. In the official spark-rapids docs I came across this page which states:
There are cases where you may want to get access to the raw data on the GPU, preferably without copying it. One use case for this is exporting the data to an ML framework after doing feature extraction.
To me this sounds as if one could make data that is already available on the GPU from some upstream Spark ETL process directly available to a framework such as Tensorflow or PyTorch. If this is the case how can I access the data from within any of these frameworks? If I am misunderstanding something here, what is the quote exactly referring to?
The link you references really only allows you to get access to the data still sitting on the GPU, but using that data in another framework, like Tensorflow or PyTorch is not that simple.
TL;DR; Unless you have a library explicitly setup to work with the RAPIDS accelerator you probably want to run your ETL with RAPIDS, then save it, and launch a new job to train your models using that data.
There are still a number of issues that you would need to solve. We have worked on these in the case of XGBoost, but it has not been something that we have tried to tackle for Tensorflow or PyTorch yet.
The big issues are
Getting the data to the correct process. Even if the data is on the GPU, because of security, it is tied to a given user process. PyTorch and Tensorflow generally run as python processes and not in the same JVM that Spark is running in. This means that the data has to be sent to the other process. There are several ways to do this, but it is non-trivial to try and do it as a zero-copy operation.
The format of the data is not what Tensorflow or PyTorch want. The data for RAPIDs is in an arrow compatible format. Tensorflow and PyTorch have APIs for importing data in standard formats from the CPU, but it might take a bit of work to get the data into a format that the frameworks want and to find an API to let you pull it in directly from the GPU.
Sharing GPU resources. Spark only recently added in support for scheduling GPUs. Prior to that people would just launch a single spark task per executor and a single python process so that the python process would own the entire GPU when doing training or inference. With the RAPIDS accelerator the GPU is not free any more and you need a way to share the resources. RMM provides some of this if both libraries are updated to use it and they are in the same process, but in the case of Pytorch and and Tensoflow they are typically in python processes so figuring out how to share the GPU is hard.
Related
I realize this is not the intended usage of TensorRT, but I am a bit stuck so maybe there are some ideas out there. Currently I have been provided some neural network models as TensorRT serialized engines, so-called .trt files. These are basically models compiled and optimized from PyTorch to run on a specific GPU.
Now, this works fine since I do have a compatible GPU for development, however, for setting up CI/CD, I am having some trouble because the cloud servers on which it will be running for testing purposes only do not have adequate GPUs for this CUDA-compiled "engine".
So, I would like to force these models to run on CPU, or otherwise find some other way to make them run. On CPU would be just fine, because I just need to run handful of inferences to check the output, it is fine if it's slow. Again, I know this is not the intended usage of TensorRT, but I need some output from the models for integration testing.
Alternative approach
The other idea I had was maybe to convert the .trt files back to .onnx or another format that I could load into another runtime engine, or just into PyTorch or TensorFlow, but I cannot find any TensorRT tools that load an engine and write a model file. Presumably because it is "compiled" and no longer convertible; yet, the model parameters must be in there, so does anyone know how to do such a thing?
Currently I am using Python, Numpy, pandas, scikit-learn to do data preprocessing (LabelEncoder, MinMaxScaler, fillna, etc.), and then feeding the processed data to DNN models built with Tensorflow 2.0. This input pipeline meets my needs when data is small enough to fit a PC's RAM.
Now I have some large datasets, more than 10GB, some are larger. I also plan to deploy the models in a production environment, which means there will be new data coming everyday. For DNN model training there is distributed strategy of tensorflow 2.0. But for data preprocessing obviously I cannot use pandas, scikitlearn on the large datasets with one PC. It seems to me I need to use a for-loop where I repeatedly fetch a small part of the data and use it for training?
I am wondering what do people typically use in either experiment or production environment for big data preprocessing?
Should I use Spark(Scala) / PySpark and Tensorflow input pipeline?
Yeah, with the current way you are doing preprocessing, it'll not scale well.
PySpark is one right way to run your preprocessing layer. Setup a simple standalone spark cluster with few workers and then run your preprocessing (labelEncoder/OneHotEncoder/fillNA/...) This solution should scale well and it abstracts the distributed computation layer.
PS : PySpark might not be the only way forward, but it is one of the good way forward for this use case.
I have programmed some code doing an inference with Tensorflow's C API (CPU only). It is running on a cluster node, where I have access to 24 CPUs and 1 GPU. I do not make use of the GPU as I will need to do the task CPU-only later on.
Somehow every time I call the Tensorflow-Code from the other program (OpenFOAM) Tensorflow seems to run on all CPUs parallelized. However I have not done anything to cause this behavior. Now I would like to know whether Tensorflow does this parallelization by default?
Greets and thanks in advance!
I am not sure how you are using tensorflow. But a typical TensorFlow training has an input pipeline which can be thought as an ETL process. Following are the main activities involved:
Extract: Read data from persistent storage
Transform: Use CPU cores to parse and perform preprocessing operations on the data such as image decompression, data augmentation transformations (such as random crop, flips, and color distortions), shuffling, and batching.
Load: Load the transformed data onto the accelerator device(s) (for example, GPU(s) or TPU(s)) that execute the machine learning model.
CPUs are generally used during the data transformation. During the transformation, the data input elements are preprocessed. To improve the performance of the pre-processing, it is parallelized across multiple CPU cores by default.
Tensorflow provides the tf.data API which offers the tf.data.Dataset.map transformation. To control the parallelism, the map provides the num_parallel_calls argument.
Read more on this from here:
https://www.tensorflow.org/guide/performance/datasets
I have a system with two GPUs, and am using Keras with Tensorflow backend. Gpu:0 is being allocated to PyCUDA, which is performing a unique operation which is fed forward to Keras, and changes with each batch. As such, I would like to run a Keras model on gpu:1 while leaving gpu:0 allocated to PyCUDA.
Is there any way to do this? Looking through prior threads I've found several depreciated solutions.
So I don't think that this feature is meaningfully implemented in Keras currently. Found a workaround that I recommend whereby you just create multiple processes using Python's default multiprocessing library.
Note: Currently for this setup you need to spawn the new process, rather than fork it, to avoid a weird interaction with one of the PyCUDA backend libraries.
I'm trying to create ML models dealing with big datasets. My question is more related to the preprocessing of these big datasets. In this sense, I'd like to know what are the differences between doing the preprocessing with Dataprep, Dataproc or Tensorflow.
Any help would be appreciated.
Those are 3 different things, you can't really compare them.
Dataprep - data service for visually exploring, cleaning, and
preparing structured and unstructured data for analysis
In other words, if you have a large training data and you want to clean it up, visualize etc. google dataprep enables you to do that easily.
Cloud Dataproc is a fast, easy-to-use, fully-managed cloud service for
running Apache Spark and Apache Hadoop clusters in a simpler, more
cost-efficient way.
Within the context of your question, after you cleanup your data and it is ready to feed into your ML algorithm, you can use Cloud Dataproc to distribute it across multiple nodes and process it much faster. In some machine learning algorithms the disk read speed might be a bottleneck so it could greatly improve your machine learning algorithms running time.
Finally Tensorflow:
TensorFlow™ is an open source software library for numerical
computation using data flow graphs. Nodes in the graph represent
mathematical operations, while the graph edges represent the
multidimensional data arrays (tensors) communicated between them.
So after your data is ready to process; you can use Tensorflow to implement machine learning algorithms. Tensorflow is a python library so it is relatively easy to pick up. Tensorflow also enables to run your algorithms on GPU instead of CPU and (recently) also on Google Cloud TPUs(hardware made specifically for machine learning, even better performance than GPUs).
In the context of preprocessing for Machine Learning, I would like to put a time to answer this question in details. So, please bear with me!
Google provides four different processing products. Since, preprocessing has different aspects and covers many different ML prerequisites, each of these platforms is more suitable for a particular preprocessing domain. Products are as follows:
Google ML Engine/ Cloud AI: This product is based on Tensorflow. You can run your Machine Learning code in Tensorflow on the ML Engine. For specific types of data like image, text or sequential, tf.keras.preprocessing or tf.contrib.learn.preprocessing Libraries are available to make the appropriate input/tensor format of data for Tensorflow rapidly.
You may also need to transform your data via tf.Transform in a preprocessing step. tf.Transform, a library for TensorFlow, allows users to define preprocessing pipelines as part of a TensorFlow graph. tf.Transform ensures that no skew can arise during preprocessing.
Cloud DataPrep: Preprocessing sometimes is defined as data cleaning, data cleansing, data prepping and data alteration. For this purposes, Cloud DataPrep is the best option. For instance, if you want to get rid of null values or some ASCII characters which may cause errors in your ML model, you can use Cloud DataPrep.
Cloud DataFlow, Cloud Dataproc: Feature extraction, feature selection, scaling, dimension reduction also can be considered as a part of ML preprocessing. Since Cloud DataFlow and DataProc both support Spark, one can use Spark libraries for distributed fast preprocessing of the ML models input. Apache Spark MLlib can also be applied to many ML preprocessing/processing. Note that since Cloud DataFlow supports Apache Beam, it is more into stream processing while Cloud DataProc is more Hadoop-based and is better for batch preprocessing. For more details, please refer to Using Apache Spark with TensorFlow document