Agglomerative Clustering with Python: Dendrogram, Heat Map

Agglomerative Clustering: Start with each item as an individual unit, and then create clusters by combining clusters together.

Types:
* Single linkage: use minimum distance.
* Complete linkage: use maximum distance.
* Average linkage: use average distance.
* Centroid linkage: distance between group means.
* Ward's Method: considers loss of information between individual records and group mean.

A dendrogram is a nice way to cluster, and view clusters. A dendrogram gives a visual indication of clusters merging.

In this video, I create a dendrogram of Ohio counties.

I import the Ohio county data, and then compute distances. After that, I noke the dendrogram function to generate a dendrogram.

To validate clusters, we consider:
- Cluster interpretability. Is the interpretation reasonable?
* Obtain summary statistics from each cluster
* Examine clusters for separation among common feature that was not used in analysis.
* Label the clusters

- Stability: do clusters change if some of the inputs are altered?
* Partition the data to see how well clusters form based on the two parts.

- Cluster separation: the ratio of variation between cluster to variation within the cluster
- Number of clusters

Next, I create a heatmap to give a visualization to the factors that were considered in the cluster. We can create a heatmap with sns.clustermap() function. This also gives us a dendrogram.

A few limitations:
- It's computationally expensive for large datasets.
- Algortihm makes one pass at the data, so a mistake early cannot be reallocated.
- Low stability: dropping records can lead to a different solution.
- Sensitive to outliers.

Видео Agglomerative Clustering with Python: Dendrogram, Heat Map канала Brandan Jones