Clustering Algorithms - Science topic

Dear Emil

running K-means Clustering several times is a good first approach.

Taking the "best" result in the end is a fine strategy.

But consider to widen the approach from just comparing "best" results on the summed and squared differences between data and clusters to some quality criterion like:

Sihouette Coefficient, Gap-Statistics, Calinski-Harabasz index, Dunn index, Davies-Bouldin index, and others that have been mentioned above

and then decide to use one of those, and explicitly optimize all meta parameters (like no of patterns, initial K, ...) w.r.t this index.

For image compression i had fruitful results using the Calinski-Harabasz index.

regards

Nils

Луценко Е.В., Подсистема агломеративной когнитивной кластеризации классов системы «Эйдос» ("Эйдос-кластер"). Пат. № 2012610135 РФ. Заяв. № 2011617962 РФ 26.10.2011. Опубл. От 10.01.2012. – Режим доступа: http://lc.kubagro.ru/aidos/2012610135.jpg, 3,125 у.п.л.

Луценко Е.В. Метод когнитивной кластеризации или кластеризация на основе знаний (кластеризация в системно-когнитивном анализе и интеллектуальной системе «Эйдос») / Е.В. Луценко, В.Е. Коржаков // Политематический сетевой электронный научный журнал Кубанского государственного аграрного университета (Научный журнал КубГАУ) [Электронный ресурс]. – Краснодар: КубГАУ, 2011. – №07(071). С. 528 – 576. – Шифр Информрегистра: 0421100012\0253, IDA [article ID]: 0711107040. – Режим доступа: http://ej.kubagro.ru/2011/07/pdf/40.pdf, 3,062 у.п.л.

Thank you. As mentioned in the question, I am using Structure (version 2.3).

Have you managed to implement it?

Jaya Sharma

Thank you again dear Inès François

I didn't use sequence plot previously.. have you tried that?

I am asking because i want to know if i can extract the used colours in this plot ( which will be the aimed states in my case) .. or i have to re-cluster the images of these plots and utilize from get-colour function to identify the used colours ..

Actually, i thought about that since i might end up in more than 5 time series clusters ..

So, in this case does sequence plots use same colour for the number that will be plotted? or each plot will be populated with different colour for each number?

Angana Ray Yep, you are accurate. Cd-hit-est employs a word counting method to group sequences with a high degree of similarity depending on a user-specified identity threshold. The identity criterion is set to 90% by default, which indicates that sequences with 90% or higher similarity will be grouped together. In your example, Seq1 and Seq2 have no nucleotide sequence similarity, thus they are clustered together as a singleton cluster with Seq1 as the "typical" sequence and a similarity score of -/100%. This score denotes that the sequence bears no resemblance to any other sequence in the cluster.

I think clustering is one way of classifying. Therefore, after clustering, some additional classification is not required.

Myles Joshua Toledo Tan Converting a.csv file to an AnnData file for use in Scanpy is a simple procedure. Here's an example of how it may be done:

1. You must first install the anndata package, which can be done by typing pip install anndata into your command line.

2. Following that, import the relevant libraries, such as pandas and anndata.

import pandas as pd

import anndata

3. Then, using the pd.read csv() method, read your.csv file into a pandas DataFrame.

data = pd.read_csv("your_file.csv")

4. After that, you can use the anndata.AnnData() method to convert the DataFrame to an AnnData object.

adata = anndata.AnnData(data)

5. Finally, you may use the scanpy library's different methods to conduct any extra processing or analysis on your AnnData object.

It's worth noting that when you convert a dataframe to an AnnData object, it thinks the rows are observations and the columns are variables. If you have the inverse, use.T to transpose the dataframe.

You can alternatively use scanpy.read csv() to directly import the csv file into an AnnData object to get the same result.

adata = sc.read_csv("your_file.csv")

Please let me know if there is anything else I can do for you.

You can check this publication where I use MLP (DL). Adagrad give good result .

@article{islam2022explainable, title={Explainable machine learning methods for classification of brain states during visual perception}, author={Islam, Robiul and Andreev, Andrey V and Shusharina, Natalia N and Hramov, Alexander E}, journal={Mathematics}, volume={10}, number={15}, pages={2819}, year={2022}, publisher={MDPI} }

K-MDTSC: K-Multi-Dimensional Time-Series Clustering Algorithm.

If you are asking for Brain rs-fMRI,

1. Neuronal activation patterns between brain regions

2. dynamic Functional Connectivity(dFC) network to record the temporal changes

Yet another Python implementation:

Dear Fady Samann ,

PAM stands for “partition around medoids”. The algorithm is intended to find a sequence of objects called medoids that are centrally located in clusters.

Regards,

Shafagat

Select representative features using cluster firstly and then use the optimization algorithm to fit and solve the classification problem using the selected features

Hi,

Depends on the your algorithm and your goal, there may be lots of datasets to work on. Besides, you can take any image classification dataset (like MNIST) and simply ignore its labels, then set your goals to put images with the same label in a same group. Also, I find these two datasets a good starting point.

Hello Ahmed Ashlam,

I certainly avoid answering inquiries if I have vague notions regarding a particular topic. I always read a question before responding so that people do not think I am an imbecile.

Hi Imad,

Here are three links on the subject that will interest you, I think the first link provides a thesis, despite its date 2010 remains current.

Communication among UAVs Thesis in Computer Engineering Jun 2010

Chaoyou Dai Yifei Li Weiming Zhai

link.springer.com/chapter/10.1007%2F978-3-319-13823-7_25 -

Here is a non-exhaustive list of tools can be used as UTSim, FlyNetSim, UAV Toolbox MathWorks including Matlab…

For more details about this subject i suggest you to see links and attached flile on topic.

Article Unmanned Aerial Vehicle Propagation Datalink Tool Based on a...

Best regards

Thank you for taking the time to reply Muhammad Ali .

I looked at the linked resources but do not see anything related to my question (*nested categorical* variables, not simple categorical variables). In case I missed it, could you please indicate the relevant section?

Kind regards,

Mick

Muhammad Ali Thank you for your response. I couldn't find a link on unsupervised clustering, though!

Thank you Shafagat Mahmudova for your cooperation.

Roberto Mancebo Dimension reduction is crucial in cluster analysis since it results in less data storage while retaining the same analytical outcomes as the original representation. To achieve an efficient processing time when clustering and to reduce the curse of dimensionality, a clustering algorithm requires data reduction.

Also, take a look at:

Maybe one idea is to use character embedding instead of word embedding. Higher-level embeddings also could be useful such as morpheme level or unsupervised subword tokenization.

Thanks Aravinda C V , however, I insist that I am looking for a script that provides a way to perform a cluster analysis between "raster files"

Dear Tong Guo

In the following paper,

a new embedding technique based on deep learning for text clustering has been proposed.

Dear Amin,

This is a good question. The below link describes very well the solid definition of transfer learning and mentioned why and when we should use a pre-trained model

"A pre-trained model may not be 100% accurate in your application, but it saves huge efforts required to re-invent the wheel."

Umair Munir

Did you find anything that can help me, too?

Dear Cecilia,

Hope these links can help you!

the boundary becomes smoother with increasing value of K. The training error rate and the validation error rate are two parameters we need to access different K-value.

To get the optimal value of K, you can segregate the training and validation from the initial dataset. Now plot the validation error curve to get the optimal value of K. This value of K should be used for all predictions.

The optimal K value usually found is the square root of N, where N is the total number of samples. Use an error plot or accuracy plot to find the most favorable K value. KNN performs well with multi-label classes, but you must be aware of the outliers.

There is no solution for your problem. Either change your clustering strategy or run the process on super/cluster computer which has terrabytes of RAM.

If you do a classification task you could just use both classification algorithm k-means and CNN to classify then you'll be more confident about your classification (even better if you use more than just two methods)

Dear Rahul Gaikwad

Have a quick with these papers also:

1. https://www.semanticscholar.org/paper/Performance-analysis-of-Data-Mining-algorithms-in-Tiwari-Jha/758d8e910cb39750c241e92df8847d35c2df84a4

2. https://www.ijera.com/papers/vol%201%20issue%203/I013441445.pdf

3. https://www.journalijar.com/article/18041/performance-evaluation-of-weka-clustering-algorithms-on-large-datasets/

Kind Regards

Hi Muhammad Zahid Khattak , I have this kind of data.

Write me and we will discuss about the possibility to make some experiments together.

Simply, you may make some values to empty.

I typically have used k-means clustering algorithm which is very popular. This algorithm is based on partitioning. Similarly you can use clustering algorithms based on density or hierarchical clustering methods.

I'm not keen on calling anything "smart". Any method will fail under some circumstances, such as for some outlier that no-one have thought of.

Cross Recurrence or Joint Recurrence are the mot general metrics for correlating time series. The first is phhase dependent the other not.

see:

The number of parameters, the values for the training parameters, the size and topology of the map, all have to be determined in advance. However, there are several settings to bring good results but they are time-consuming.

Dear Zainab Mostafa

I do have that code with Python PL.

If your interested, please keep me posted.

Regards .

This is for homogeneous environment, what will be for heterogeneous environment?

Convolutional Neural Networks (CNNs) is the most popular neural network model being used for image classification problem. The big idea behind CNNs is that a local understanding of an image is good enough

Top 5 Classification Algorithms in Machine Learning

You may try this algorithm:

B. K. Tripathy and D. Mittal: Hadoop based uncertain possibilistic kernelized c-means algorithms for image segmentation and a comparative analysis, Applied soft computing, 46, (2016), pp.886-923.

Yes, you can use a different number, It will produce a different outcome when compared to using 42, which can be used to evaluate your experiment in distinct scenarios.

Also, you are probably using '42' because of this (from wikipedia): The number 42 is, in The Hitchhiker's Guide to the Galaxy by Douglas Adams, the "Answer to the Ultimate Question of Life, the Universe, and Everything" =)

Sara Mirzaie: I suggested one way in my second answer above. I'd be happy to discuss further via email or a direct message here :-)

Performance/accuracy comparison is based on the similar task over similar data using different method.

Have a look at

https://www.researchgate.net/post/Is_there_any_clustering_method_for_clustering_users_by_their_features_eg_their_profile_properties2

https://archive.siam.org/meetings/sdm11/clustering.pdf

Have a look at some related work, such as : http://users.wpi.edu/~yli15/courses/DS504Spring18/includes/denclue.pdf

https://towardsdatascience.com/density-based-algorithms-49237773c73b

Hello Prabhakar Eswaran

These followings are the top search engine, need proper keywords, (and keep changing to reach suitable keywords):

1. https://datasetsearch.research.google.com/

2. https://oatd.org/

3. https://www.usgs.gov/

4. https://www.nodc.noaa.gov/access/

Also, there are many more such as https://data.mendeley.com/

Parva Jain

Please find the answer below.

1- Your question:

Answer

Technically, you need to build a model that works on the selected feature subset. To do so, a search method like "BestFirst" must be applied to find the feature subset. Then, you need to apply an evaluator to assess the merit of the selected subset. Finally, invoke cluster algorithm like EM (expectation maximisation) to find the number fo cluster (Note that finding the number of clusters is not a trivial task by default, ut EM does this job using internal cross-validation.)

2- Your question:

actually the user will pick his preferred attributes and I want to apply clustering for all the possible preferences a user can have.

Answer

If you have a user to select specific features, then no need to have a search method, just apply EM directly on the selected features.

HTH.

Dr. Samer Sarsam

In addition to the well-written paper that Sakib Shahriar suggested you could also have a look at that one:

This paper would be supportive:

Dear Mohsen,

Have a look at http://www.mit.edu/~9.54/fall14/slides/Class13.pdf

Yes, to make sure that your calculations will not be biased either to the very high or to the very low values. In other words, to make sure that all your data are at the same level. you could use any normalization technique to do this, and I recommend this:

X_i(new)= (X_i-mean(all X values))/standard deviation

For K-means, data Normalization is not always required (however it always improves clustering), but it rarely hurts. Some examples: K-means: K-means clustering is "isotropic" in all directions of space and therefore tends to produce more or less round (rather than elongated) clusters. See also : https://www.researchgate.net/post/Does_normalization_of_data_always_improve_the_clustering_results

Hello Davide

Do you have interest on feture selection or hyper parameter optimization?

regards

Dear author

please check below science site:

please recommend our response

Thank you very much

Good Luck

Dear

i suggest K-mean

use Cross-validation

I would recommend to also look into Boolean matrix factorization (BMF). k-means/modes is only usable if clusters have convex shapes and every point belongs to exactly one cluster. BMF can compute clusters with overlap (might not be necessarily important for your application) but it also identifies the features which are important for creating one cluster together with the data points belonging to one cluster. Especially if you have a high-dimensional feature space, it is unlikely that the similarity between points is expressed over the whole feature space. BMF identifies the feature space in which the points form a cluster.

If you are interested, look at Chapter 2 (particularly Section 2.4) from

Dear Ayaz Hussain ,

Look the link, maybe useful.

Regards,

Shafagat

Try to figure out is there any methods in which you can find a correlation or some other relationship among the heterogeneous outputs of the different clustering algorithms, then you can use the ensemble approach.

You can evaluate the performance of k-means by convergence rate and by the sum of squared error(SSE), making the comparison among SSE.

read about it more at url

https://towardsdatascience.com/k-means-clustering-algorithm-applications-evaluation-methods-and-drawbacks-aa03e644b48a

Dear Seyyed Mohammad Mahdi Bahadornia ,

Let’s think in a practical use of DBSCAN. Suppose we have an e-commerce and we want to improve our sales by recommending relevant products to our customers. We don’t know exactly what our customers are looking for but based on a data set we can predict and recommend a relevant product to a specific customer. We can apply the DBSCAN to our data set (based on the e-commerce database) and find clusters based on the products that the users have bought. Using this clusters we can find similarities between customers, for example, the customer A have bought 1 pen, 1 book and 1 scissors and the customer B have bought 1 book and 1 scissors, then we can recommend 1 pen to the customer B. This is just a little example of use of DBSCAN, but it can be used in a lot of applications in several areas.

https://forums.ni.com/t5/Example-Program-Drafts/Artificial-Neural-Network-Multilayer-Perceptron-with-Back/ta-p/3491335?profile.language=esIn the NI forums some examples can be found of the application you're asking for. these two papers who invented backpropagation.Geoffrey hinton - https://www.cs.toronto.edu/~hinton/absps/plautfilters.pdfYann lecunn - http://yann.lecun.com/exdb/publis/pdf/lecun-88.pdf

Please see

Possibly, (random) graphs is what would help you

Fuzzy clustering is better than k-means clustering because it gives the more appropriate results. You can go through one of my paper which is applied on text data.

Anton N. Gvozdetsky , Shafagat Mahmudova , Pratyay Sengupta thank you all so much!

Regards,

Antonio

Hi, you can you Approximate Computing (AC) to further minimize the running time of such algorithms for big data and save computational time and energy in the process. You can check out the different tools and technologies available for AC.

I have a survey paper on the same. You can have a look as well.

INDIVIDUAL DIFFERENCES.

Thank you Basim Mahmood

Micro- cluster is a temporal extension of the cluster feature, which compresses the data effectively. It defines the clusters by separating dense area from sparse ones. In clustering data streams, it is impractical to save all the incoming data objects. Micro-clusters are a popular technique in stream clustering, which maintain the compact representation of the clustering.

The following link may be useful for you.

http://www.iaeng.org/publication/IMECS2011/IMECS2011_pp410-414.pdf

sunupo@126.com。Thanks for your kindness

Nguyen Van Thieu thank you so much!! your answer is really useful.

The centroid of each data labeled group correspond to the cluster prototipe