Intel® oneAPI Hackathon for Open Innovation

Environment set up for various run:

Step 1: Open the condaenvsetup.ipynb and activate Python 3(Intel oneAPI 2022.3) kernel
Step 2: Run through the cells one by one. "qsub" submits the job to do all the setup and "qstat" shows the job status. (all details related to devcloud job submission is in the Welcome.ipyb file)
Step 3: Once the job is finished, the setup should be completed.
Step 4: Now you will have stock-tensorflow as well as intel-tensorflow in the jupyter kernels to run the benchmark code.
Step 5: You will also have intel-modin kernel to run the benchmark code.

Running the notebooks:

Exercise 1: Intel® extension for Pytorch

Step 1: Open terminal and clone oneAPI samples
git clone https://github.com/oneapi-src/oneAPI-samples.git
step 2: Browse oneAPI-samples/AI-and-Analytics/Features-and-Functionality/IntelPyTorch_Extensions_Inference_Optimization/
step 3: Open the optimize_pytorch_models_with_ipex.ipynb
step 4: Activate the Pytorch(AI kit) kernel.
step 5: Run all the cells for to see comparision between baseline pytorch and IPEX optimization features for inference.

Exercise 2: Intel® Tensorflow and Stock Tensorflow Performance Comparison

Step 1: Open terminal, and clone oneAPI samples
git clone https://github.com/oneapi-src/oneAPI-samples.git
Step 2: Browse the oneAPI-samples/AI-and-Analytics/Features-and-Functionality/IntelTensorFlow_ModelZoo_Inference_with_FP32_Int8/
Step 3: Open the ResNet50_Inference.ipynb
Step 4: First select stock-tensorflow kernel and add the following env variable [os.environ["TF_ENABLE_ONEDNN_OPTS"]='0'] to disable onednn and run through each cell of the notebook to get the average time and throughput.
Step 5: Follow step 4 but now with onednn on by changing the env variable [os.environ["TF_ENABLE_ONEDNN_OPTS"]='1'] and note the performance
Step 6: Repeat the same process but this time with intel-tensorflow kernel to add further openmp optimizations and compare the average time and throughput

Exercise 3: Intel® Extension for Scikit-learn

Step 1: Browse the oneAPI-samples/AI-and-Analytics/Features-and-Functionality/Intel_Extension_For_SKLearn_Performance_SVC_Adult/
Step 2: Open the Intel_Extension_for_SKLearn_Performance_SVC_Adult.ipynb
Step 3: Activate the python3(Intel oneAPI 2022.3) kernel. Run through all the cells. This compares both the Intel extension SKLearn with the Unoptimized SKLearn and compares the performance optimization.

Exercise 4: Intel® Distribution for Modin

Step 1: Browse /oneAPI-samples/AI-and-Analytics/Getting-Started-Samples/IntelModin_GettingStarted/
Step 2: Open IntelModin_GettingStarted.ipynb and activate the intel-modin kernel.
Step 3: Run all the cells which has modin and pandas comparision.

Exercise 5: XGBoost Optimized for Intel®

Step 1: Browse /oneAPI-samples/AI-and-Analytics/Features-and-Functionality/IntelPython_XGBoost_Performance/
Step 2: Open IntelPython_XGBoost_Performance.ipynb
Step 3: First activate Python 3(Intel oneAPI 2022.3) kernel and run till 8th cell.
Step 4: Check if perf_numbers.csv file has been created.
Step 5: Now install xgboost=0.81 inside juypter notebook to change the xgboost version.
Step 6: Run all the cell again to see performace comparision.