TensorFlow

Units of data are called tensors. Tensors are a set of primative values (represented by an array/vector)

• rank is the dimensionality of a tensor (array)
• tensorflow is modelled with a computaitonal graph
  • series of operations represented as a graph
• simplist node is a constant, taking zero inputs, and outputting a stored constant
• this function abstractly represents a constant, but isn't one until it runs in a session
• tf.Session.run(tensor1) would return the function output of tensor1
• tf.Variable([...], tf.float32 ) constructs a variable with an initial value and type
• variables are not initalized in a tf program until tf.global_variables_initializer()
• operations are nodes, tf.add, used to combine tensors
• tf.placeholder(tf.float32) are a promises to accept external input values (for user parameterization)
• tf.assign(myVar, myVal) can be passed to the tf session to reassign variable values

MNIST is a computer vision dataset, and a simple task is to use a simple model called softmax regression -- source

• one-hot vectors are 0 in most dimensions, and 1 in a single dimension
  • for MNIST digits, [0,0,0,1,0,0,0,0,0,0] represents the literal number 3
• goal is to assign a probability to each possible output (0..9), given an image input
• softmax regression is used to assign probabilities to each output, acting as a simple model itself or a final layer in a more complex system