Let's take a look at an example to see how hard an OR problem can be!

This video is about showing the boundaries of mathematical optimization.

What are different mathematical approaches?

What are their limitations and advantages?

A higher level overview on Operations Research.

An example of MILP formulation to get a better sense on how MILP solves the problems.

How can we formulate an OR problem and solve it?

A higher level overview of how many algorithms work to get to know how Dynamic programming and Reinforcement Learning algorithms work.

Delve into the common natures of RL algorithms.

What is Dynamic Programming?

Let's solve a DP problem by hands-on calculations.

What is Markov Decision Process and how it related to RL?

Let's talk about the heart of DP and RL algorithms!

Now it's time for rap up everything and introduce a framework that DP and RL models rely on.

What are the components of Reinforcement Learning framework?

A first attempt on solving RL problems!

Let's get to know one of the famous algorithms in RL which is a foundation for many other algorithms as well.

What is off-policy and on-policy and what are their roles in RL?

Finally, let's talk about Deep RL and its similarities and differences with tabular RL.

Let's talk about the google Colab that we will code there.

What is the resource allocation problem?

Let's define parameters related to the problem.

Let's define parameters related to DP algorithm.

Transition matrix is the heart of DP. Let's define it.

Transition matrix is the heart of DP. Let's continue defining it.

Let's introduce the framework for solving DP problems which is the foundation for many other algorithms in RL as well.

First phase of GPI algorithm.

Second Phase of GPI algorithm!

Let's interpret the obtained results.

Let's introduce inventory optimization problem.

Define the parameters for the problem and algorithm.

Let's code action selection mechanism.

What will be the reward for inventory optimization problem?

Let's code the Bellman equation for Q-Learning.

How should we interpret the results?

What is Travel Sales Person (TSP) problem?

Define parameters for problem and solution.

Epsilon-greedy action selection.

A function to update Q-table.

Let's frame a main structure of Q-learning algorithm.

Let's extract the best route out of Q-table.

How can we design our own customized environments based on OpenAI gym library?

Again, Let's define the parameters for the last time :)

How should we define reset() function in the environment?

Let's define step() function, as a core of all RL environments and let's see some tips on how to do it properly?

Let's write a test to see whether the environment works properly or not!

Let's write a Deep Learning network to serve as our q-value approximator.

What is Reply buffer?

Let's define reply buffer!

Let's rap everything up and create a DQN agent that works. Let's initialize the parameters and variables that we need.

Let's define Epsilon-Decayed-Greedy action selection policy!

Why we need two networks to have a stable learning process?

Let's get the Q-values from policy network.

Let's update policy network parameters.

Let's update target network.

Visualization of loss values show us how the agent is learning. Let's do it!

AND finally, let's get the best route from the trained agent.

Let's write the main loop for DQN agent training!