Python Tutorial: Basics of cluster analysis скачать в хорошем качестве

Python Tutorial: Basics of cluster analysis

Want to learn more? Take the full course at https://learn.datacamp.com/courses/cl... at your own pace. More than a video, you'll learn hands-on coding & quickly apply skills to your daily work. --- Hello everyone! Now that you are familiar with unsupervised learning, let us move on to the basics of cluster analysis. First, let us define a cluster. It is a group of items with similar characteristics. In the Google News example, a cluster of news articles has similar words and word associations appearing in them. Another example of clustering is segmentation of customers based on their spending habits. A cluster of customers would exhibit similar spending habits. Although there are other algorithms like the density based DBSCAN technique or Gaussian models for cluster analysis, this course will focus only on the two most common techniques - hierarchical and k-means clustering. Let us look at the steps in hierarchical clustering before the implementation. For demonstration purposes, let us consider 13 points on a 2D plane. In the first step, all the points are considered as individual clusters. A cluster center is a mean of attributes of all data points in a cluster. In this case, cluster centers will have two attributes - the mean of x and y coordinates. At this stage, cluster centers of all clusters are the coordinates of the individual points. Next, the distances between all pairs of cluster centers are computed and the two closest clusters are merged. The cluster center of the merged cluster is then recomputed. In the example, two clusters on the bottom left have been merged. At this step, we are left with 12 clusters, one less than we started. In the second step, the clusters with the closest cluster centers are merged on the top left. This process of merging the two closest clusters continues until we arrive at the desired clusters. At every step, the number of clusters reduces by one. As visible from the distribution of points, let us stop at three clusters. Finally, these are three clusters that you arrive at after the algorithm has run. Let us now look at its simple implementation in SciPy. The required methods for hierarchical clustering, linkage and fcluster, are stored in scipy-dot-cluster-dot-hierarchy. We will explore the parameters of these methods later in the course. We additionally import pyplot and seaborn for visualization and pandas for data manipulation. A list of fifteen points with x and y coordinates is stored in a dataframe. First, the linkage method computes distances between intermediate clusters. Next, the fcluster method generates clusters and assigns associated cluster labels to a new column in the dataframe. Finally, we plot the points using seaborn with the cluster labels as the hue argument to associate clusters with different colors. Here is the result of the seaborn scatterplot, showing the three clusters. Next, we discuss the algorithm of k-means clustering with the same set of points to create three clusters. First, a random cluster center is generated for each of the three clusters. Next, the distance to these cluster centers is computed for each point to assign to the closest cluster. The cluster centers are recomputed. This iteration of assigning points to the recomputed cluster centers is performed a predefined number of times. Here, the clusters have been formed in the first iteration. K-means clustering is implemented using methods kmeans and vq of scipy-dot-cluster-dot-vq. The centroids of the clusters are computed using kmeans and cluster assignments for each point are done through vq. The second argument in both methods is distortion, which we capture in a dummy variable. We will explore the parameters of these methods later in the course. Finally, we plot the clusters using seaborn. Here is the resulting plot. Let us try some exercises now. #DataCamp #PythonTutorial #Cluster #Analysis #Python