Yang Yang(Tony) tonyyang-svail

Python Notes

from __future__ import print_function
import os
import psutil

process = psutil.Process(os.getpid())

Main inspiration comes from [here][1].

“”” Here is what a deep learning system stack would look like in nowdays.

Build operator level graph description language: name whatever dl frameworks you care about, and [ONNX][2]
Tensor primitive level graph description languages: [NNVM][3], [HLO/XLA][4], [NGraph][5]. It is close enough to the first one that you can also build graph optimization on first layer and bypass this layer.
DSL for description and codegen: TVM, image processing languages like [halide][6], [darkroom][7].
Hardcoded optimized kernel library: [nnpack][8], [cudnn][9], [libdnn][10]
Device dependent library: [maxas][11](assembler for NVIDIA Maxwell architecture)

Start a pod:

kubectl run $POD_NAME --image=$IMAGE_NAME --port=$PORT_NUMBER --image-pull-policy=Never

Get stdout of POD/CONTAINER

kubectl logs two-ubuntu-bash bash1

Originally developed at LinkedIn. Written in Java. Client module supports many languages.

Kafka = A Distribued Commit Log

	class MyClass(object):
	pass

	# is identical to

	# type(name, bases, dct)
	# - name is a string giving the name of the class to be constructed
	# - bases is a tuple giving the parent classes of the class to be constructed
	# - dct is a dictionary of the attributes and methods of the class to be constructed
	MyClass = type('MyClass', (), {})

	Despite the characteristic of a programming language, compiled or scripted,
	static typing or dynamic typing, the large scale software project written in
	that language is essentailly a tree of files.

	When coding in C, and many other compilation languages, the following things
	are done separately

	- Building System: a practical building system (e.g. Makefile, CMake,
	Bazel) does not come with the language itself, one need to install one.
	Extra code has to be written.

	from __future__ import absolute_import
	from __future__ import division
	from __future__ import print_function

	import tensorflow as tf

	import tensorflow.contrib.eager as tfe
	tfe.enable_eager_execution()

	NOISE_DIMENSION = 10

	from __future__ import absolute_import
	from __future__ import division
	from __future__ import print_function

	import tensorflow as tf

	import tensorflow.contrib.eager as tfe
	tfe.enable_eager_execution()

	INPUT_DIMENSION = 784

	with parameters() as params:
	fc1 = layers.Dense(hidden_dim, input_shape=(input_dim,))
	fc2 = layers.Dense(output_dim, input_shape=(hidden_dim,))

	def forward(images, labels):
	x = fc1(images)
	x = layers.relu(x)
	x = fc2(x)
	logits = layers.relu(x)
	loss = losses.softmax_cross_entropy(logits, labels)