I’m trying to plot kmeans results and getting irregular results I don’t understand. Some insight / advice from someone more experienced would be greatly appreciated.
Here is my code.
kmeans = KMeans(n_clusters=8, init=‘k-means++’, max_iter=300, n_init=10, random_state=0)
pred_y = kmeans.fit_predict(df_scaled)
cluster_new = df.copy()
cluster_new[“cluster_pred”] = kmeans_new.fit_predict(df_scaled)
First plot
plt.figure(figsize=(10,7))
plt.scatter(df_scaled[:,0], df_scaled[:,1], s=20, c = cluster_new[“cluster_pred”], cmap=“rainbow”)
plt.scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1], s=300, color=‘red’)
plt.title(‘K Means Cluster Analysis - Banknote Authentication’)
plt.xlabel(“V1”)
plt.ylabel(“V2”)
plt.show()
Plot at top
This looks good, however, the data are standardized (df_scaled) and the centroid centers plot well, however, the coordinates for some of the cluster centroids should not plot within the limits of the axes displayed, so they must be scaled to the initial scatterplot.
First several rows of the df_scaled data
Sample of “df_scaled” data
[[ 1.12180565 1.14945512]
[ 1.44706568 1.06445293]
[ 1.20780971 -0.77735215]
…
[-1.47235682 -2.62164576]
[-1.40669251 -1.75647104]
[-1.04712236 -0.43982168]]
If I try to plot the data using the data arising from kmeans analysis - cluster_new[“cluster_pred”] = kmeans_new.fit_predict(df_scaled), the data from with slightly greater range in the axes, consistent with the data, however, the cluster centroids appears to have been scaled again. Visual examination of the centroid coordinates with respect the plot shows they are not plotting at their proper coordinates (i.e., Group 0 - x = -7.26279682e-01, y = -2.23211523e-01)
plt.figure(figsize=(10,7))
plt.scatter(cluster_new[‘V1’], cluster_new[‘V2’], s=20, c = cluster_new[“cluster_pred”], cmap=“rainbow”)
plt.scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1], s=300, color=‘red’)
plt.title(‘K Means Cluster Analysis - Banknote Authentication’)
plt.xlabel(“V1”)
plt.ylabel(“V2”)
plt.show()
Sample Cluster_new data
V1 V2 cluster_pred
0 3.62160 8.66610 1
1 4.54590 8.16740 1
2 3.86600 -2.63830 3
3 3.45660 9.52280 1
4 0.32924 -4.45520 3
… … … …
1367 0.40614 1.34920 2
1368 -1.38870 -4.87730 4
1369 -3.75030 -13.45860 4
1370 -3.56370 -8.38270 4
1371 -2.54190 -0.65804 0
[1372 rows x 3 columns]
Plot of bottom of previous figure
How can I get the cluster centroids to plot correctly with respect to the coordinates from the cluster_new file?