Have any question ? +91 8106-920-029 +91 6301-939-583 team@appliedaicourse.
com My Account Logout
Live Sessions
Home Courses Applied Machine Learning Online Course Revision Questions Search Videos
Revision Questions COMPLETE
Instructor: Applied AI Course Duration: 30 mins
Revision:
What is dimensionality reduction? https://www.appliedaicourse.com/lecture/11/applied- Plotting for exploratory data analysis
(EDA)
machine-learning-online-course/2878/what-is-dimensionality-reduction/2/module-2-data-
science-exploratory-data-analysis-and-data-visualization
Explain Principal Component Analysis?https://www.appliedaicourse.com/lecture/11/applied- Linear Algebra
machine-learning-online-course/2889/geometric-intuition-of-pca/2/module-2-data-science-
exploratory-data-analysis-and-data-visualization
Probability and Statistics
Importance of PCA?https://www.appliedaicourse.com/lecture/11/applied-machine-learning-
online-course/2888/why-learn-pca/2/module-2-data-science-exploratory-data-analysis-and-
data-visualization
Interview Questions on Probability
and statistics
Limitations of PCA?https://www.appliedaicourse.com/lecture/11/applied-machine-learning-
online-course/2894/limitations-of-pca/2/module-2-data-science-exploratory-data-analysis-
and-data-visualization
Dimensionality reduction and
Visualization:
What is t-SNE?https://www.appliedaicourse.com/lecture/11/applied-machine-learning-online-
course/2898/what-is-t-sne/2/module-2-data-science-exploratory-data-analysis-and-data-
visualization
PCA(principal component analysis)
What is Crowding problem?https://www.appliedaicourse.com/lecture/11/applied-machine-
learning-online-course/2901/crowding-problem/2/module-2-data-science-exploratory-data-
analysis-and-data-visualization (t-SNE)T-distributed Stochastic
Neighbourhood Embedding
How to apply t-SNE and interpret its output?
https://www.appliedaicourse.com/lecture/11/applied-machine-learning-online-
7.1 What is t-SNE?
course/2902/how-to-apply-t-sne-and-interpret-its-output/2/module-2-data-science- 7 min
exploratory-data-analysis-and-data-visualization
7.2 Neighborhood of a point,
Embedding 7 min
Prev Next
Code example of t-SNE Questions & Answers 7.3 Geometric intuition of t-SNE
9 min
25 Comment(s)
search comments Search 7.4 Crowding Problem
8 min
Leave a response
7.5 How to apply t-SNE and interpret
its output 38 min
Format 7.6 t-SNE on MNIST
7 min
7.7 Code example of t-SNE
9 min
7.8 Revision Questions 30 min
Interview Questions on Dimensionality
Reduction
Module 2: Live Sessions
Submit
Sujit Jena 7 Votes
What is Dimensionality Reduction?
In machine learning classification problems, there are often too many factors on the basis of
which the final classification is done. These factors are basically variables called features. The
higher the number of features, the harder it gets to visualize the training set and then work
on it. Sometimes, most of these features are correlated, and hence redundant. This is where
dimensionality reduction algorithms come into play. Dimensionality reduction is the process of
reducing the number of random variables under consideration, by obtaining a set of principal
variables. It can be divided into feature selection and feature extraction.
Reply May 14, 2019 14:04 PM
AppliedAI Team
what is doubt here?
Reply May 14, 2019 20:41 PM
Sujit Jena
I have shared this as an answer , let me know if I am wrong
Reply May 14, 2019 20:44 PM
AppliedAI Team
That is right.
Reply May 14, 2019 21:34 PM
Ramya Vidiyala 6 Votes
Here are my answers. Please let me know if anything is wrong or needs to be added.
1 Dimensionality reduction means projecting data to a lower-dimensional
space, which makes it easier for the visualization and analysis of data.
2 Explain Principal Component Analysis?
PCA means finding out the components (features) which are effective to the data and
discarding the redundant features.
3 Importance of PCA?
With few lines of codes we can reduce the dimensions by a huge number.
4 Limitations of PCA?
PCA does preserve the global direction of the data but not the local, which creates confusion
when an overlap of 2 clusters happens after the reduction.
5 What is t-SNE?
t-SNE stands for t-distribution Scholastic Neighbourhood Embedding.
Scholastic – not definite but random probability
Neighborhood – concerned only about retaining the structure of neighborhood points.
Embedding – plotting data into lower dimensions tSNE is the state of the art or one of the
best techniques for dimensionality reduction, which is widely used for data visualization.
6 What is Crowding problem?
When a data point, ‘x’ is a neighbor to 2 data points that are not neighboring to each other,
this may result in losing the neighborhood of ‘x' with one of the data points as t-SNE is
concerned only within the neighborhood zone.
7 How to apply t-SNE and interpret its output?
· There are 3 parameters
a) Steps: number of iterations
b) Perplexity: can be thought of as the number of neighboring points.
c) Epsilon: It is for data visualization and determines the speed which it should be changed.
· Points to remember while performing tSNE are
1. Never stop with a single-step value. Check for various values and take the value
at which the plot is stable.
2. With lower perplexity values, we may see a few shapes of clusters. But do not
fall into the trap. Try with various Perplexity values ranging from 2 to the number of data
points. But, remember a value of 2 or a value equal to a number of data points will lead to no
information.
3. Never come to any conclusions with random data.
4. As tSNE is Scholastic, each run may lead to
slightly different. However, by setting random_state, this can be solved.
5. tSNE doesn’t preserve the distance between clusters. So, when we have
multiple clusters, we might not retain the similar distance between the
clusters.
6. tSNE shrinks the widespread data and expands densely packed data. So based
on the output, we cannot decide on the cluster size and density/ spread /
variance of the clusters.
Reply May 16, 2020 15:55 PM
team aaic
Your understanding is correct. You can refer this blog for better understanding of PCA
importance.
Reply May 16, 2020 18:47 PM
Slim Shady
which blog?
Reply Jun 27, 2020 21:05 PM
team aaic
Please refer this blog : https://towardsdatascience.com/pca-clearly-explained-how-
when-why-to-use-it-and-feature-importance-a-guide-in-python-7c274582c37e
Reply Jun 28, 2020 10:53 AM
Chinda Mani Teja Verma 6 Votes
give some pdf document to the above questions.
Reply Feb 01, 2019 12:58 PM
Applied AI Course
Sure, will work on it.
Reply Feb 02, 2019 08:29 AM
Debasish Acharya
can we get the pdf?
Reply Jun 07, 2020 01:02 AM
Applied_AI
Please drop us a mail at teamappliedaicourse.com
Reply Jun 07, 2020 12:57 PM
Ayush Agarwal 3 Votes
Can you please provide questions like "What change would you do in your model if there is
this problem or that,"?
Reply Mar 02, 2019 12:51 PM
AppliedAI Team
thanks for your feedback. We will definitely try to create some context based interview
question as you suggested and will update these. We have given some questions here.
Reply Mar 03, 2019 11:54 AM
Sunney Sood
the link given above does not work..please check and repost. Thanks
Reply Apr 16, 2019 14:06 PM
Sadiva Madaan 2 Votes
Dimensionality Reduction means projecting a data matrix from a higher dimension to a lower
dimension . It basically removes all the redundant features from our data .
PCA helps in finding the most important features . With the help of it we can find eigen values
and eigen vectors. They help us to know by how much we have to rotate our axis for
maximum variance / information .
Importance of PCA - We can massively reduce the dimensions of our data matrix with a few
lines of code.
Limitations of PCA - PCA doesn't work well when we have our data distributed in the form of
a circle or in clusters . It preserves the global shape but fails to preserve the local shape .
t-SNE - t - Students t distribution, S - Stochastic (It means that it is not deterministic but is
probabilistic ), N - Neighborhood (t-SNE main objective is to preserve the structure of
neighborhood points ), E - Embedding (It means picking up a point from high dimensional
space and placing it into lower dimension).
Sometimes it is impossible to preserve the distances in all the neighborhoods . This problem
is called Crowding Problem .
How to apply t-SNE -
1) There are two most important parameters - No of steps and Perplexity .
Perplexity means the no of neighborhood points to be preserved . Steps means the no of
iterations it should perform .
2) Always run t-SNE with multiple perplexity values .
3) If perplexity = no of data points then it will create a mess.
4) t-SNE never actually replicates the data .
Reply Aug 12, 2020 22:19 PM
team aaic
Great summary of both concepts(PCA and t-SNE). Thanks for sharing.
Reply Aug 12, 2020 22:28 PM
Karthik 1 Votes
How can we find the best perplexity value ? (Like elbow method of k-means)
Reply Nov 16, 2019 23:27 PM
appliedai course
we can fix a large steps like steps=5000 and try out various values of p and take that p as
best p when the model stabilises(i.e model does not change as we increase p)
Reply Nov 17, 2019 05:01 AM
Subrahmanyam Kesani
As we increase no. of iterations, we will reach a stage where the result will not
change further. Is it the same case with perplexity also ? I mean... after certain
perplexity...won't the result change anymore by increasing perplexity (using
"random_state=0") ? In an example shown in the video "How to apply t-SNE and
interpret its output ?", the stabilized cluster points become disarray when perplexity
is increased further.
t-SNE undoubtedly is giving beautiful clusters we wish for. But I am doubtful on
when to conclude based on perplexity.
Reply May 11, 2020 16:19 PM
appliedai course
no if we increase perplexity further cluster points disarray. refer this under
'those hyperparameters really matter" section for constant steps of 5000,
perplexity of 30 and 50 are stable but for perplexity 100 results messed up
Reply May 11, 2020 18:58 PM
Abhinaya Srinivasan 0 Votes
Why does the visualization of the data becomes effective or better when the number of
iterations increases in case of a T-SNE plot?
Reply Jun 23, 2020 11:04 AM
Applied AI Tech Admin
It is because when the number of iterations increase, the more neighborhood will be
preserved and at a particular number of iterations, we see the shape of the points to be
constant.
Reply Jun 23, 2020 12:13 PM
Lohit Krishna 0 Votes
I was doing with another which has stings as the value of the feature. Will TSNE not allow
strings?
Do we need to drop the column before using TSNE.
Please let me know so that I can proceed further.
Another question what are the dimensions that TSNE is bringing up to us in form of DIM1 and
DIM2.
Can I use PCA in combination with TSNE that is I will computer the Eigen vector values and
USE TSNE for the eigen vectors because I think that would even fetch us accurate results and
even dimensionality reduction will also be in a perfect manner as eigen vectors are computed
Reply Nov 29, 2018 06:13 AM
Applied AI Course Team1
1. For any machine learning model, we need to give numbers. We have to convert the
data that we have into numbers.
2. Those are just two dimensions. The data that we have is just a 2D representation of
higher dimensional data.
3. You can use truncated svd, as it is a general case SVD. You can use that to reduce it to
some dimensions based on the explained variance and then apply t-sne. And yes. it will
give better results.
Reply Nov 29, 2018 11:08 AM
Previous 1 Next
Our courses Contact us More
Applied Machine Learning Course +91 8106-920-029 Success stories
AI/Machine Learning Case Studies +91 6301-939-583 Job Guarantee
AI Workshop (whatsapp business) Live Sessions
Desktop Application
Student Blogs
Terms & Conditions
© 2019- All rights are reserved- AAIC Technologies pvt ltd
