Neural Networks Google Professional Data Engineer GCP

In this, we will learn about the concepts of Neural Networks.

It has multiple nodes that interconnects to each other,
mimics the neurons in our brain.
Each neuron inputs from another neuron, performs tasks, and transfers to another as output.

Neural Networks In figure

Each circular node is an artificial neuron
arrow represents a connection from the output of one neuron to the input of another.

As per diagram, there are 4 layers present – Layer 1, Layer 2, Layer 3 and Layer 4.
Every deep neural network consists of three types of layers, which are:
Input Layer (Layer 1): provides the input parameters and passes parameters to further layers without any computation at this layer.
Hidden Layers (Layers 2 and 3): Perform the computations on the inputs receiving from the previous layers and pass on the result to the next layer. More the number of hidden layers, deeper is the network.
Output Layer (Layer 4): gives final output after receiving the results from the previous layers.

This attaches some weightage to a certain feature.
Some features might be more important than others
weights calculate the weighted sum for each neuron. x1, x2, x3, x4 represent the weights associated with the corresponding connections in the deep neural network.

Backpropagation

This is for neural net training.
For fine-tuning weights of a neural net based on the error rate obtained in the previous epoch (i.e., iteration).
Proper tuning of weights reduce error rates and make model reliable.
Backpropagation is a short form for “backward propagation of errors.”
standard method of training artificial neural networks.
calculate the gradient of a loss function with respects to all the weights in the network.

Inputs X, arrive through the preconnected path
Input is modeled using real weights W, randomly selected.
Calculate the output for every neuron from the input layer, to the hidden layers, to the output layer.
Calculate the error in the outputs: ErrorB= Actual Output – Desired Output
Travel back from the output layer to the hidden layer to adjust the weights such that the error is decreased.
Keep repeating the process until the desired output is achieved

Backpropagation Networks Types:

Static back-propagation: produces mapping of a static input for static output.
Recurrent Backpropagation: It is fed forward until a fixed value is achieved. After that, the error is computed and propagated backward.

should be reasonably related
Feature value should be known at the time of prediction without latency
Numeric with range of meaningful magnitude.
Difficult for categorical inputs.
Need to find vector/bitmask representation(1 hot encoding), for NL, we can use wordtovec.
Need to have enough examples of each feature value.
For continuous numbers, use windows.
Feature crosses can be done using human sights, this will simplify ml model.

Linear models work well for sparse features(like employee id).
Neural network model works well for dense features(like price, color(rgb))
Regressor – linear, sparse features, specific, many features, memorization, employee id, nlp
DNN – non-linear, dense features, generalization, color, price
Wide and linear – combination of both, recommendation, search and ranking

Changing model parameters to find the right set of parameters for a particular result.
evaluate rmse with the parameter and keep tuning to lower rmse.
define what metric we want for evaluation like rmse.

A type of Machine Learning where machine is required to determine the ideal behaviour within a specific context, in order to maximize its rewards.
It works on the rewards and punishment principle
for any decision which a machine takes, it will be either be rewarded for correct or punished.
So machine will learn to take the correct decisions to maximize the reward in the long run.
Can focus on long-term rewards or the short-term rewards.
If machine is in a specific state and has to be the action for the next state in order to achieve the reward, it is called the Markov Decision Process.
The environment is modelled as a stochastic finite state machine with
- inputs (actions sent from the agent)
- outputs (observations and rewards sent to the agent)
It requires clever exploration mechanisms.
Selection of actions with careful reference to the probability of an event
It is memory expensive to store the values of each state

Agent: It is an assumed entity which performs actions in an environment to gain some reward.
Environment (e): A scenario that an agent has to face.
Reward (R): An immediate return given to an agent when he or she performs specific action or task.
State (s): State refers to the current situation returned by the environment.
Policy (π): It is a strategy which applies by the agent to decide the next action based on the current state.
Value (V): It is expected long-term return with discount, as compared to the short-term reward.
Value Function: It specifies the value of a state that is the total amount of reward. It is an agent which should be expected beginning from that state.
Model of the environment: This mimics the behavior of the environment. It helps you to make inferences to be made and also determine how the environment will behave.
Model based methods: It is a method for solving reinforcement learning problems which use model-based methods.
Q value or action value (Q): Q value is quite similar to value. The only difference between the two is that it takes an additional parameter as a current action.

cat is an agent that is exposed to the environment. In this case, it is house. An example of a state could be cat sitting, and you use a specific word in for cat to walk.
Our agent reacts by performing an action transition from one “state” to another “state.”
For example, cat goes from sitting to walking.
The reaction of an agent is an action, and the policy is a method of selecting an action given a state in expectation of better outcomes.
After the transition, they may get a reward or penalty in return.

Value-Based: try to maximize a value function V(s).
Policy-based: try to come up with such a policy that the action performed in every state helps you to gain maximum reward in the future.
Model-Based: need to create a virtual model for each environment.