TensorFlow How to Start

tf-how-to-start.md

TensorFlow How to Start

Free Cloud Computational Power (Powerful enough for education without local setup hassle):

• https://colab.research.google.com/
• Tutorial: https://medium.com/deep-learning-turkey/google-colab-free-gpu-tutorial-e113627b9f5d

Local Setup:

Setup Anaconda 3: https://docs.anaconda.com/anaconda/install/

I recommend to setup environment with Python 3.6: conda create -n tensorflow python=3.6 anaconda Once you use python 3.6, do not upgrade or downgrade it. If you want to use different python version, create different environment. I learn this in a hard way.

Setup Jupyter Notebook:

• https://jupyter.org/install, use the pip install

Add conda environment to Jupyter Notebook:

https://medium.com/@nrk25693/how-to-add-your-conda-environment-to-your-jupyter-notebook-in-just-4-steps-abeab8b8d084

Setup TensorFlow in Python

Current tensorflow stable version is 2.3.1, which has a lot if improvement over 2.1.0 and some breaking changes. A lot of things working now without workaround.

Setup TensorFlow CPU (Relative Easy):

• https://www.tensorflow.org/install/pip, code: pip install tensorflow==2.3.1

Setup TensorFlow GPU (Currently only support NVidia GPU):

• https://blog.quantinsti.com/install-tensorflow-gpu/, code: pip install tensorflow-gpu==2.3.1
• https://www.youtube.com/watch?v=tPq6NIboLSc

Setup TensorFlowJS Converter

• https://www.tensorflow.org/js/tutorials/conversion/import_keras
• Code: pip install tensorflowjs
• TF Converter Wizard, tensorflowjs_wizard, and follow the instruction.

TensorFlow JS

We can train tensorflow model in JavaScript, or deploy the model to web for production use. Current tfjs version is 2.0.0 and has breaking change with 1.7.x. But the best thing with this 2.0.0 we can choose the backend between webgl, cpu or wasm. And wasm is the faster for smaller size model.

TensorFlow JS Code Examples:

• https://github.com/tensorflow/tfjs-examples/
• Face and hand tracking in the browser with MediaPipe and TensorFlow.js: https://blog.tensorflow.org/2020/03/face-and-hand-tracking-in-browser-with-mediapipe-and-tensorflowjs.html

TensorFlow JS Node (if you want to use nodejs to do the training):

• https://www.npmjs.com/package/@tensorflow/tfjs-node
• or https://www.npmjs.com/package/@tensorflow/tfjs-node-gpu

Free Online course:

1. Intro to TensorFlow for Deep Learning: https://classroom.udacity.com/courses/ud187/
2. Machine Learning Foundations: https://www.youtube.com/playlist?list=PLOU2XLYxmsII9mzQ-Xxug4l2o04JBrkLV
3. Machine Learning Recipes with Josh Gordon: https://www.youtube.com/playlist?list=PLOU2XLYxmsIIuiBfYad6rFYQU_jL2ryal
4. Coding TensorFlow: https://www.youtube.com/playlist?list=PLQY2H8rRoyvwLbzbnKJ59NkZvQAW9wLbx
5. Natural Language Processing (NLP) Zero to Hero: https://www.youtube.com/playlist?list=PLQY2H8rRoyvzDbLUZkbudP-MFQZwNmU4S

Datasets and lots of sample:

• Kaggle: https://www.kaggle.com/

My Code Samples and Presentation:

• Serving TensorFlow Models in various ways: https://docs.google.com/presentation/d/1aV816OgCMKaJr9eOTNc1uFpFI2d96zKsUpXYj9ZZDDU/edit?usp=sharing
• Coin Object Detection: https://github.com/ivankristianto/coincalculator
• Implementtion TFLite with same Coin Object Detection model: https://github.com/ivankristianto/coin-detector-android
• Dog breed classification: https://github.com/ivankristianto/tf-dogbreed-classification
• Predict Pneumonia from Chest x-ray: https://github.com/ivankristianto/xray-pneumonia

Resources:

• TF Homepage: https://www.tensorflow.org/
• TF Python API: https://www.tensorflow.org/api_docs/python/
• TF Hub: https://tfhub.dev/
• Keras Homepage: https://keras.io/
• Keras API Doc: https://keras.io/api/