Nimbix execution steps and sample exercise
- 1. Steps to Run Jupyter Notebook on Nimbix and Sample DL Usecase OpenPOWER Academia
- 2. Execution Below are the steps to run jupyter notebook on IBM PowerAI over nimbix cloud Login to the nimbix platform https://platform.jarvice.com/ Click on Compute and Search for IBM PowerAI in the search box
- 3. Execution Click on the IBM PowerAI, to get it expanded Click on Launch Button and select the machine type as per the requirement.
- 4. Execution Click on Submit button to start the machine. Click on the Blue area desktop to login to the machine, it will open up a new browser.
- 5. Execution Open Terminal Emulater for shell and type below command to activate Tensorflow, you can activate any DL framework supported. source /opt/DL/tensorflow/bin/tensorflow-activate
- 6. Execution Once Tensorflow activated start Jupyter notebook by typing ipython notebook, then click on new Notebook.
- 7. Execution In order to persist data, Nimbix provides File Manager to store the data even after VM shuts down, this data is available from VM root directory under data folder To start File Manager, click on compute and search for File Manager, then open up “Nimbix File Manager” icon then start it as we did for Power AI. Once started click on Connect to open it in new browser Copy the data folder inside nibmix File Manager inside data folder.
- 8. OCR Tensorflow Optical Character Recognition using Tesorflow Exercise
- 9. Introduction Objective is to do a Optical Character Recognition vis Deep learning approach using Tensorflow This Demo is the execution of the work done by Nicholas Smith https://nicholastsmith.wordpress.com/2017/10/14/deep-learning-ocr-using- tensorflow-and-python/@ https://github.com/tensorflow/models/tree/master/research/object_detection This program generate set of 32K OCR images as training dataset then it trains the models based in that dataset. It also generates 256 Test images too for testing.
- 10. Execution Steps Login to nimbix platform. Launch PowerAI VM with minimal configuration available. Connect the VM via putty client with host and credential displayed on VM start in nimbix This demo requires python 3.5 and Tensorflow 1.5, by default PowerAI VMs do not provide this combination it still runs on python 2.7,we need to install setup python 3.5 and Tensorflow 1.5 and will not use the DL framework provides by Power AI by default Follow this blog to make the above environment ready https://developer.ibm.com/code/howtos/install-tensorflow-on-power Once tensorflow got set up clone the repository https://github.com/nicholastoddsmith/pythonml.git Code uses the font Consolas.ttf for image generation which is not provided by default in Unix for that clone the fonts https://github.com/kakkoyun/linux.files.git in /home/nimbix/pythonml/DeepOCR
- 11. Execution Steps Edit OCRGen.py with this path TF = ImageFont.truetype('linux.files/fonts/Consolas.ttf', 18) Create Trn and Tst directory inside /home/nimbix/pythonml/DeepOCR. Install below dependencies sudo apt-get update sudo apt-get install libatlas-base-dev pip3 install scikit-image (Will take 15-20 mins) pip3 install sklearn pip3 install TFANN Run python3 OCRGen.py, this will generate 32K Training Image and 256 Testing images in Trn and Tst folders respectively. There is a pretrained model already created inside TFModel we can use that for our testing To convert images from Tst directory execute python3 DeepOCR.py Tst/00000.png This is will display the converted text on console. Code is written with specific image dimension, we need to tweak this code to make it generic.
- 12. Subscribe to OpenPOWER Academia and Research Channel https://www.youtube.com/channel/UCYLtbUp0AH 0ZAv5mNut1Kcg