Lets start with “Tensor Flow” open-sourced by Google ….for Machine learning
What is TensorFlow ? you must have heard of TF, if not, lets learn about TF :) I have just started learning TF.
TF is a library, used for designing and running ML & Deep learning.
We can create our own Neural network using TF and we write our TF programs in Python.
TF has its own datatype called “Tensor”, same as NumPy library.
TF is based on lazy computing, same as Functional programming.
We can create Tensor objects [Array/Pandas/List]. Once the Tensor object is created, the computational Graph is created automatically.
The TF graph contains multiple default operations(start, stop, etc).
We need to use Session object to actually run the Graph (pass input values to Tensor). Session is the runtime in which the Tensor graph operations will run.
As a developer, we need to build the Graph or use the default Graph.
Benefits of use Tensor Flow:
- use GPU’s
- Distributed processing
- Inspiration from Theano and Torch
- Tensor processing is faster than NumPy — numerical processing
How to build TensorFlow :
- Placeholders =Num of Examples →(Features:inputs and Targets:outputs)
- Parameters and Operations= (m, b) : (Weights, Biases) = (Y = mX+b)
- Cost = what is the error rate? comparing the predicted output with Actual output
- Optimization = specify the Learning Rate and this will minimize Cost
- Train=Process in which Algorithm is learning from Input & creating a New Model, which will be used for Prediction
TF will keep updating the Parameters (m, b) in multiple Steps to minimize the Cost. TF will try to reduce the Cost in sequence of Steps by changing the Parameters, so that the accuracy rate % is high close to 90+%.
At every Step, TF will change the Parameters (m,b).
Please watch this Video: