To define our kingdoms, let's draw a line between our sand castles! Don't worry we can draw a curved line if necessary!
Supportive Horses and Flying Dogs! The aim for the SVN is to classify between two groups of data sets and draw a line between them. This is possible to find a subset of data for each group and determine the 'border' and classify two groups of datasets. The subset of the data that is uses here are supportive data points and hence from my kingdom a subset of horses, and from your kingdom the subset of flying dogs (apologies for my drawings!).
Gift and Curse of SVN [consider buffer zones!]. As you noticed by now, we are looking at the subset of data points to classify. What does it mean for us? Well, it means it takes less computation time but there is no magic here, because we are not considering all data points overlapping occurs. , so some degree of tolerance you need to think about, so that you tolerate overlapping points (aka classificaiton errors).
Cost Parameter. We can create the buffer zones by using cost parameter where the degree of tolerance is determined. And to increase the tolerance level you need to have a bigger cost parameter. To what degree to push the cost parameter for accurate prediction, we need to use cross validation, that we will cover in its own post in the near future.
Summary. To find the optimal separation between two datasets with the given dataset use SVN. If you have small datasets, want to classify more than between two groups, and when you need to have a very large tolerance for classification to happen (this means you have too many overlapping data points), do not use SVN.
Artificial Intelligence Machine Learning Support Vector Machine SVN Classification PredictionPosted by Seyed Hossein Ghotbi (Published on April 10, 2022).
Let's predict the pinky rabbit among other pinky rabbits, or detect a strange blue one!
Who has neighbours?! K-NN stands for K-nearest neighbours, and therefore even if we don't know what K is at this point, we know that its process is based on the close by neighbours just by its name alone. So if we think for a minute, we realise, oh a dataset that consists of a set of group-by data points (clusters) could be an input dataset for the K-NN.
Let's go one step further, a cluster of data points have similar values (that's why they are clustered after all!), so if I know the values of all the neighbours of a data point in the cluster, then it makes sense that somehow I could also predict the value of the unknown point. This is where K-NN comes in: it classifies the data points by majority of neighbours, and tries to predict the value.
How many neighbours do we need? Aha!, that is the question here as well as the meaning of K!. K means how many number of nearest neighbours we need to look at to predict the value?. This process is actually called parameter tuning. Which we cover will in future posts.
What if the rabbit is blue among all pink rabbits? So if we can predict the value of a data point within a group of data, then we should also be able to detect anomalies as well, an this is part of the predictive models' (such as K-NN) usage. Just a quick note here is that, the data size does matter in this case, because if our data size is small, then you have less data points to work with and that causes anomalies.
Artificial Intelligence Machine Learning K-nearest neighbours K-NN Predictive Analysis Anomaly DetectionPosted by Seyed Hossein Ghotbi (Published on April 2, 2022).
You want a human be in the loop? That is the question!
The creation of rules, and how this creation is done is the core difference between machine learning algorithms that uses different processes for learning. Today, we look at two important ones: supervised and unsupervised learning.
Unsupervised Learning. to put it simply, the learning cycle on the given data is not interrupted ed by humans. To look at this the other way: we don't know what needs to be learned within the data for prediction, but however we want the unsupervised algorithm to find it. For example, when you go to an online store as a customer, and want to buy an item, the algorithm in the back of the online store could predict what is the likelihood of you buying the item based on your past purchases and the time of the year. This prediction happened by unsupervised learning algorithms that learn from all the purchase history of customers, and could come with the prediction.
Supervised Learning. here we as humans interrupt the learning by giving the machines ideas what to predict. For example, in our datasets we label the pictures of cats as 'cat' and dogs as 'dog' and then let the machine learn what is cat and what is dog, so in future a new image comes in the machine could predict if it is a cat or a dog.
Artificial Intelligence Machine Learning Supervised Learning Unsupervised Learning Supervised Vs. Unsupervised LearningPosted by Seyed Hossein Ghotbi (Published on March 27, 2022).
PyTorch will be imported in your Python code, so we need to write our PyTorch code in Python. Therefore first we need to check if we got
Python 3 in our system. Simply open the Terminal, and typing the following:
. Here we get the following: Python 3.9.1. If you don't have Python 3, there is an official beginner guide for its installation on your system.
Step 2: We install NumPy & Matplotlib
Pytorch uses tensors (will cover this later, don't worry!) that are multi-dimensional arrays, and when we want to visualise them we convert them into Numpy arrays and by using matplotlib library we can visualise them. So let's install them both in our system.
pip3 install numpy matplot lib
Note: We are using pip3, and our installation is global on the system.
Pytorch is a deep learning framework that help us to develop deep learning models for our machine learning projects and products. For installation on your Mac, use the following command:
pip3 install torch torchvision torchaudio
Note: In case later you want to use PyTorch with C++ or Java (No CUDA could be used), head over to PyTorch's installation documentation page.
For coding environment we use Jupyter notebook, assuming that you want to get into PyTorch right away and experiment with it without struggling too much on choosing the light/heavy IDE side of things.
brew install jupyter
Lets first fire up the Jupter notebook, so lets call it in the terminal:
New Notebook (local):Then create a new notebook in Jupyter, by clicking on new on the top right corner and create a new Python 3 notebook.
Import the installed components. We installed Pytorch, NumPy and Matplotlib, lets import them in the first cell. Type the following and then do shift + Enter , to run the first cell.
import numpy as np
import matplotlib.pyplot as plt
Artificial Intelligence Machine Learning Deep Learning PyTorch Jupyter Notebook NumPy MatplotlibPosted by Seyed Hossein Ghotbi (Published on March 26, 2022).