Getting to know TensorFlow 

Unlike conventional Python libraries, TensorFlow first builds an empty graph containing the structure of model and then uses Session to run the graph by feeding in the data. For feeding in the data, TensorFlow uses tensors. A tensor is nothing but a multi-dimensional array. Tensor is the fundamental unit of data in TensorFlow. A tensor's dimension is represented by its rank and shape is represented like a numpy matrix ([2, 3], for example).