Sampling: Definition

Sampling is defined as the process of selecting certain members or a subset of the

population to make statistical inferences from them and to estimate characteristics of
the whole population. Sampling is widely used by researchers in market research so
that they do not need to research the entire population to collect actionable insights. It is
also a time-convenient and a cost-effective method and hence forms the basis of
any research design.
For example, if a drug manufacturer would like to research the adverse side effects of a
drug on the population of the country, it is close to impossible to be able to conduct a
research study that involves everyone. In this case, the researcher decides a sample of
people from each demographic and then conducts the research on them which gives
them an indicative feedback on the behavior of the drug on the population.
Types of Sampling: Sampling Methods
Any market research study requires two essential types of sampling. They are:
1. Probability Sampling: Probability sampling s a sampling method that selects random
members of a population by setting a few selection criteria. These selection
parameters allow every member to have the equal opportunities to be a part of
various samples.
2. Non-probability Sampling: Non probability sampling method is reliant on a
researcher’s ability to select members at random. This sampling method is not a fixed
or pre-defined selection process which makes it difficult for all elements of a
population to have equal opportunities to be included in a sample.
In this blog, we discuss the various probability and non-probability sampling methods
that can be implemented in any market research study.
Types of Sampling: Probability Sampling Methods
Probability Sampling is a sampling technique in which sample from a larger population
are chosen using a method based on the theory of probability. This sampling method
considers every member of the population and forms samples on the basis of a fixed
process. For example, in a population of 1000 members, each of these members will
have 1/1000 chances of being selected to be a part of a sample. It gets rid of bias in the
population and gives a fair chance to all members to be included in the sample.
There are 4 types of probability sampling technique:
 Simple Random Sampling: One of the best probability sampling techniques that
helps in saving time and resources, is the Simple Random Sampling method. It is a
trustworthy method of obtaining information where every single member of a
population is chosen randomly, merely by chance and each individual has the exact
same probability of being chosen to be a part of a sample.
For example, in an organization of 500 employees, if the HR team decides on
conducting team building activities, it is highly likely that they would prefer picking
chits out of a bowl. In this case, each of the 500 employees has an equal opportunity
of being selected.
 Cluster Sampling: Cluster sampling is a method where the researchers divide the
entire population into sections or clusters that represent a population. Clusters are
identified and included in a sample on the basis of defining demographic
parameters such as age, location, sex etc. which makes it extremely easy for a survey
creator to derive effective inference from the feedback.
For example, if the government of the United States wishes to evaluate the number of
immigrants living in the Mainland US, they can divide it into clusters on the basis of
states such as California, Texas, Florida, Massachusetts, Colorado, Hawaii etc. This
way of conducting a survey will be more effective as the results will be organized into
states and provides insightful immigration data.
 Systematic Sampling: Using systematic sampling method, members of a sample are
chosen at regular intervals of a population. It requires selection of a starting point for
the sample and sample size that can be repeated at regular intervals. This type of
 sampling method has a predefined interval and hence this sampling technique is the
least time-consuming.
For example, a researcher intends to collect a systematic sample of 500 people in a
population of 5000. Each element of the population will be numbered from 1-5000
and every 10th individual will be chosen to be a part of the sample (Total
population/ Sample Size = 5000/500 = 10).
 Stratified Random Sampling: Stratified Random sampling is a method where the
population can be divided into smaller groups, that don’t overlap but represent the
entire population together. While sampling, these groups can be organized and then
draw a sample from each group separately.
For example, a researcher looking to analyze the characteristics of people belonging
to different annual income divisions, will create strata (groups) according to annual
family income such as – Less than $20,000, $21,000 – $30,000, $31,000 to $40,000,
$41,000 to $50,000 etc. and people belonging to different income groups can be
observed to draw conclusions of which income strata have which characteristics.
Marketers can analyze which income groups to target and which ones to eliminate in
order to create a roadmap that would definitely bear fruitful results.
Use of the Probability Sampling Method
There are multiple uses of the probability sampling method. They are:
 Reduce Sample Bias: Using the probability sampling method, the bias in the sample
derived from a population is negligible to non-existent. The selection of the sample
largely depicts the understanding and the inference of the researcher. Probability
sampling leads to higher quality data collection as the population is appropriately
represented by the sample.
 Diverse Population: When the population is large and diverse, it is important to have
adequate representation so that the data is not skewed towards one demographic. For
example, if Square would like to understand the people that could their point-of-sale
devices, a survey conducted from a sample of people across US from different
industries and socio-economic backgrounds, helps.
 Create an Accurate Sample: Probability sampling helps the researchers plan and
create an accurate sample. This helps to obtain well-defined data.
Types of Sampling: Non-probability Sampling Methods
The non-probability method is a sampling method that involves a collection of
feedback on the basis of a researcher or statistician’s sample selection capabilities and
not on a fixed selection process. In most situations, output of a survey conducted with a
non-probable sample leads to skewed results, which may not totally represent the
desired target population. But, there are situations such as the preliminary stages of
research or where there are cost constraints for conducting research, where non-
probability sampling will be much more effective than the other type.
 There are 4 types of non-probability sampling

Convenience sampling: This method is dependent on the ease of access to subjects such
as surveying customers at a mall or passers-by on a busy street. It is usually termed
as convenience sampling, as it’s carried out on the basis of how easy is it for a
researcher to get in touch with the subjects. Researchers have nearly no authority over
selecting elements of the sample and it’s purely done on the basis of proximity and not
representativeness. This non-probability sampling method is used when there are time
and cost limitations in collecting feedback. In situations where there are resource
limitations such as the initial stages of research, convenience sampling is used.
For example, startups and NGOs usually conduct convenience sampling at a mall to
distribute leaflets of upcoming events or promotion of a cause – they do that by
standing at the entrance of the mall and giving out pamphlets randomly.
 Judgmental or Purposive Sampling: In judgmental or purposive sampling, the
sample is formed by the discretion of the judge purely considering the purpose of
study along with the understanding of target audience. Also known as deliberate
sampling, the participants are selected solely on the basis of research requirements
and elements who do not suffice the purpose are kept out of the sample. For instance,
when researchers want to understand the thought process of people who are
interested in studying for their master’s degree. The selection criteria will be: “Are
you interested in studying for Masters in …?” and those who respond with a “No”
will be excluded from the sample.
 Snowball sampling: Snowball sampling is a sampling method that is used in studies
which need to be carried out to understand subjects which are difficult to trace. For
example, it will be extremely challenging to survey shelterless people or illegal
 immigrants. In such cases, using the snowball theory, researchers can track a few of
that particular category to interview and results will be derived on that basis. This
sampling method is implemented in situations where the topic is highly sensitive and
not openly discussed such as conducting surveys to gather information about HIV
Aids. Not many victims will readily respond to the questions but researchers can
contact people they might know or volunteers associated with the cause to get in
touch with the victims and collect information.
 Quota sampling: In Quota sampling, selection of members in this sampling
technique happens on basis of a pre-set standard. In this case, as a sample is formed
on basis of specific attributes, the created sample will have the same attributes that
are found in the total population. It is an extremely quick method of collecting
samples.
Use of the Non-Probability Sampling Method
There are multiple uses of the non-probability sampling method. They are:
 Create a hypothesis: The non-probability sampling method is used to create a
hypothesis when limited to no prior information is available. This method helps with
immediate return of data and helps to build a base for any further research.
 Exploratory research: This sampling technique is widely used when researchers aim
at conducting qualitative research, pilot studies or exploratory research.
 Budget and time constraints: The non-probability method when there are budget
and time constraints and some preliminary data has to be collected. Since the survey
design is not rigid, it is easier to pick respondents at random and have them take
the survey or questionnaire.

Monte Carlo-
 Monte Carlo methods refer to a series of statistical methods essentially used to find
solutions to things such as computing the expected values of a function, or
integrating functions which can't be integrated analytically because they don't have a
closed-form solution.
 The Monte Carlo method is a numerical method of solving mathematical problems by
random sampling (or by the simulation of random variables).
MC methods all share the concept of using randomly drawn samples to compute a
solution to a given problem. These problems generally come in two main categories:

 Simulation: Monte Carlo or random sampling is used to run a simulation. If you

want to compute the time it will take to go from point A to point B, given some
conditions such as the chances that it will rain on your journey or that it will snow,
the chances that there will be a traffic jam, that you will have to stop on your way to
get some gas, etc. you can set these conditions at the start of your simulation and run
 the simulation 1,000 times to get an estimated time. As usual, the higher the number
of runs or trials (here 1,000), the better your estimate.
 Integration: this is a technique useful for mathematicians. MC integration though,
while not having the greatest rate of convergence to the actual solution of the integral,
can give us a way of getting a reasonably close result at a "cheaper" computational
cost.

K-armed bandit problem

In probability theory and machine learning, the multi-armed bandit
problem (sometimes called the K- or N-armed bandit problem ) is a problem in
[1] [2]

which a fixed limited set of resources must be allocated between competing

(alternative) choices in a way that maximizes their expected gain, when each choice's
properties are only partially known at the time of allocation, and may become better
understood as time passes or by allocating resources to the choice.[3][4] This is a
classic reinforcement learning problem that exemplifies the exploration–exploitation
tradeoff dilemma. The name comes from imagining a gambler at a row of slot
machines (sometimes known as "one-armed bandits"), who has to decide which
machines to play, how many times to play each machine and in which order to play
them, and whether to continue with the current machine or try a different machine. The
multi-armed bandit problem also falls into the broad category of stochastic scheduling.
In the problem, each machine provides a random reward from a probability
distribution specific to that machine, that is not known a-priori. The objective of the
gambler is to maximize the sum of rewards earned through a sequence of lever
pulls. The crucial tradeoff the gambler faces at each trial is between "exploitation" of the
machine that has the highest expected payoff and "exploration" to get
more information about the expected payoffs of the other machines. The trade-off
between exploration and exploitation is also faced in machine learning. In practice,
multi-armed bandits have been used to model problems such as managing research
projects in a large organization like a science foundation or a pharmaceutical
company. In early versions of the problem, the gambler begins with no initial
knowledge about the machines.
The multi-armed bandit problem models an agent that simultaneously attempts to
acquire new knowledge (called "exploration") and optimize their decisions based on
existing knowledge (called "exploitation"). The agent attempts to balance these
competing tasks in order to maximize their total value over the period of time
considered. There are many practical applications of the bandit model, for example:

 clinical trials investigating the effects of different experimental treatments while

minimizing patient losses,
 adaptive routing efforts for minimizing delays in a network,
 financial portfolio design

Model Based Leaning

When someone is trying to solve a real-world problem using machine learning, they
often encounter challenges:
“I am overwhelmed by the choice of machine learning methods and techniques. There’s too much
to learn!”
“I don’t know which algorithm to use or why one would be better than another for my problem.”
“My problem doesn’t seem to fit with any standard algorithm.”
Model-based machine learning helps to address all of these challenges, and makes the
process of creating effective machine learning solutions much more transparent.

The field of machine learning has seen the development of thousands of learning
algorithms. Typically, scientists choose from these algorithms to solve specific
problems. Their choices often being limited by their familiarity with these algorithms.
In this classical/traditional framework of machine learning, scientists are constrained to
making some assumptions so as to use an existing algorithm. This is in contrast to the
model-based machine learning approach which seeks to create a bespoke solution
tailored to each new problem.
The goal of MBML is “to provide a single development framework which supports the
creation of a wide range of bespoke models“. This framework emerged from an
important convergence of three key ideas:
1. the adoption of a Bayesian viewpoint,
2. the use of factor graphs (a type of a probabilistic graphical model), and
3. the application of fast, deterministic, efficient and approximate inference
algorithms.
There are 3 steps to model based machine learning, namely:
1. Describe the Model: Describe the process that generated the data using factor
graphs.
2. Condition on Observed Data: Condition the observed variables to their known
quantities.
3. Perform Inference: Perform backward reasoning to update the prior distribution
over the latent variables or parameters. In other words, calculate the posterior
probability distributions of latent variables conditioned on observed variables.
The core idea at the heart of model-based machine learning is that all
the assumptions about the problem domain are made explicit in the form of a model. In
fact, a model is just made up of this set of assumptions, expressed in a precise
mathematical form. These assumptions include the number and types of variables in the
problem domain, which variables affect each other, and what the effect of changing one
variable is on another variable.
Value iteration and policy iteration
Value iteration and policy iteration are two classic algorithms used in the field of
reinforcement learning and Markov decision processes (MDPs) to solve for an optimal
policy.
Value Iteration:
Value iteration is an iterative algorithm that aims to find the optimal value function and
policy for a given MDP. It starts with an initial estimate of the value function and
updates it iteratively until convergence. The algorithm alternates between two steps:

a. Value evaluation: In this step, the value function is updated based on the Bellman
equation, which expresses the value of a state in terms of the expected immediate
reward and the expected value of the successor states. The update equation is given by:

V(s) <- max[∑P(s'|s,a)[R(s,a,s')+γV(s')]]

Here, V(s) represents the value of state s,

P(s'|s,a) is the probability of transitioning to state s' from state s under action a,
R(s,a,s') is the immediate reward received for transitioning from state s to s' under
action a, and
γ is the discount factor that determines the importance of future rewards.

b. Policy improvement: After updating the value function, the policy is improved by
selecting the action that maximizes the expected return from each state. The policy
improvement step is given by:
π(s) <- argmax[∑P(s'|s,a)[R(s,a,s')+γV(s')]]

Here, π(s) represents the policy for state s.

The algorithm continues iterating between value evaluation and policy improvement
until the value function converges.
Policy Iteration:
Policy iteration is another iterative algorithm for finding an optimal policy in an MDP.
It consists of two main steps:

a. Policy evaluation: In this step, the value function is computed for a given policy. It
involves solving a set of linear equations called the policy evaluation equations, which
express the value of a state in terms of the expected immediate reward and the value of
the successor states under the current policy. The update equation is given by:

V(s) <- ∑P(s'|s,π(s))[R(s,π(s),s')+γV(s')]

Here, V(s) represents the value of state s,

P(s'|s,π(s)) is the probability of transitioning to state s' from state s under the current
policy π(s),
R(s,π(s),s') is the immediate reward received for transitioning from state s to s' under
the current policy, and
γ is the discount factor.

b. Policy improvement: After evaluating the current policy, a new policy is derived by
selecting the action that maximizes the expected return from each state. The policy
improvement step is the same as in value iteration:

π(s) <- argmax[ ∑P(s'|s,a)[ R(s,a,s')+γV(s') ] ]

The algorithm continues to alternate between policy evaluation and policy
improvement until the policy converges, i.e., until the policy no longer changes between
iterations.
Both value iteration and policy iteration are guaranteed to converge to an optimal
policy under certain conditions, such as having a finite state and action space. However,
the two algorithms differ in terms of the number of iterations required and the
computational complexity. Policy iteration typically converges faster but involves
solving a system of equations in the policy evaluation step, which can be
computationally expensive. Value iteration, on the other hand, directly updates the
value function and can converge with fewer iterations but may require more
computation per iteration.

Temporal Difference Learning

Temporal Difference (TD) learning is a type of reinforcement learning algorithm that
combines elements of dynamic programming and Monte Carlo methods. It is
commonly used to learn and make decisions in sequential decision-making tasks, such
as Markov Decision Processes (MDPs) and games.
TD learning is an online learning method, meaning it can learn and update its estimates
of values or policies after every time step or interaction with the environment, rather
than waiting for the end of an episode. This makes it more suitable for real-time
applications where waiting for the completion of an episode is not feasible.
The key idea behind TD learning is to use an estimate of the value function, called the
value function approximation, to update the estimates of the values based on the
observed rewards and the transitions between states. TD learning combines the current
estimate of the value function with the estimate of the value of the next state to update
the current estimate.
One of the most well-known TD learning algorithms is the TD(0) algorithm, also known
as the TD(λ) algorithm when eligibility traces are used. TD(0) uses a bootstrapping
approach, where the value estimate of the next state is used to update the current value
estimate.

The update rule for TD(0) can be expressed as follows:

V(s) ← V(s) + α [r + γV(s') - V(s)]

where V(s) is the value estimate of state s, α is the learning rate, r is the immediate
reward received after transitioning from state s to state s', γ is the discount factor that
balances the importance of immediate and future rewards, and V(s') is the value
estimate of the next state.
TD learning algorithms are known for their ability to learn online and update estimates
efficiently. They are widely used in various domains, including game playing, robotics,
and prediction tasks in general.

Deterministic and Non-deterministic Rewards and Actions

In the context of decision-making and reinforcement learning, rewards and actions can
be classified as deterministic or non-deterministic.

Deterministic Rewards:
Deterministic rewards refer to the type of rewards that have a fixed and predictable
outcome based on a particular action or state. When an agent takes a specific action in a
given state, it will always receive the same reward. The deterministic nature of rewards
simplifies the learning process for the agent since it can easily associate actions with
specific outcomes. For example, in a game where the agent receives a reward of +10 for
reaching a certain goal state, this reward is deterministic because every time the agent
reaches that state, it will receive the same reward of +10.

Non-deterministic Rewards:
Non-deterministic rewards, on the other hand, have an element of randomness or
uncertainty associated with them. When an agent takes a particular action in a given
state, it may receive different rewards each time, even if the action is repeated in the
same state. Non-deterministic rewards can introduce additional complexity to the
learning process since the agent needs to estimate the expected value or distribution of
rewards associated with different actions. An example of non-deterministic rewards
could be a game where the agent receives a reward of +5 with a certain probability, but
there is also a chance of receiving a reward of -2 or +10.

Similarly, actions can also be classified as deterministic or non-deterministic:

 Deterministic Actions:
Deterministic actions are those where the outcome or effect of taking a particular action
in a given state is known with certainty. When an agent selects a deterministic action, it
will always lead to the same resulting state. For instance, in a game where the agent can
move in four directions (up, down, left, right), if it chooses to move up from a particular
state, it will always end up in the same state.

Non-deterministic Actions:
Non-deterministic actions are actions where the outcome or effect of taking a particular
action in a given state is uncertain or probabilistic. The same action taken in the same
state can lead to different resulting states with certain probabilities. This introduces an
element of randomness in the decision-making process. For example, in a game where
the agent can move in four directions but there is a chance of slipping and moving in a
random direction, selecting the action "up" may result in moving up with a high
probability but also slipping to the left or right with a lower probability.

The classification of rewards and actions as deterministic or non-deterministic depends

on the specific problem domain and the nature of the interactions between the agent
and the environment.

Computational Learning Theory:

Computational learning theory (or just learning theory) is a subfield of artificial
intelligence devoted to studying the design and analysis of machine
learning algorithms.
In addition to performance bounds, computational learning theory studies the time
complexity and feasibility of learning. In computational learning theory, a
computation is considered feasible if it can be done in polynomial time. There are two
kinds of time complexity results:
 Positive results – Showing that a certain class of functions is learnable in polynomial
time.
 Negative results – Showing that certain classes cannot be learned in polynomial time.
Negative results often rely on commonly believed, but yet unproven assumptions such
as:
 Computational complexity – P ≠ NP (the P versus NP problem);
 Cryptographic – One-way functions exist.
There are several different approaches to computational learning theory. These
differences are based on making assumptions about the inference principles used to
generalize from limited data.
This includes different definitions of probability and different assumptions on the
generation of samples.
CLT focuses on studying “learnability,” or what functions/features are necessary to
make a given task learnable for an algorithm.
Computational learning theory has led to several practical algorithms. For example,
PAC theory inspired boosting, VC theory led to support vector machines, and Bayesian
inference led to belief networks.

Mistake bound analysis

Mistake bound analysis is a type of algorithm analysis which places bounds on the
maximum number of mistakes of an online prediction algorithm.

The sample complexity of a machine learning algorithm represents the number of

training-samples that it needs in order to successfully learn a target function.
More precisely, the sample complexity is the number of training-samples that we need
to supply to the algorithm, so that the function returned by the algorithm is within an
arbitrarily small error of the best possible function, with probability arbitrarily close to
1.
There are two variants of sample complexity:
The weak variant fixes a particular input-output distribution;
The strong variant takes the worst-case sample complexity over all input-output
distributions.

Occam learning
In computational learning theory, Occam learning is a model of algorithmic learning
where the objective of the learner is to output a concise representation of received
training data. This is closely related to probably approximately correct (PAC) learning,
where the learner is evaluated on its predictive power of a test set.
Occam learnability implies PAC learning, and for a wide variety of concept classes, the
converse is also true: PAC learnability implies Occam learnability.

VC) dimension
(VC) dimension is a measure of the capacity (complexity, expressive power, richness, or
flexibility) of a space of functions that can be learned by a statistical
classification algorithm. It is defined as the cardinality of the largest set of points that
the algorithm can shatter.