def main(argv):
X, K, init, movie_ids = DataProcessing.process_input(argv)
if init == "random":
clusters, centroids, distances = KMeans.execute(X, K)
print(
"Ran k-means. Start Distance={:.0f}, End Distance={:.0f}. Clusters = {}."
.format(distances[0], distances[-1], clusters))
Utils.write_output_csv(clusters, "output.csv", movie_ids)
elif init == "k-means++":
clusters, centroids, distances = KMeanspp.execute(X, K)
print(
"Ran k-means++. Start Distance={:.0f}, End Distance={:.0f}. Clusters = {}."
.format(distances[0], distances[-1], clusters))
Utils.write_output_csv(clusters, "output.csv", movie_ids)
elif init == "1d":
X = PCAHelper.pca_helper(X, 1)
X.astype(np.float16)
distances_by_k, cluster, centroids = OneDKmeans(
X, K).run() # KMeans.execute(X, k)
print("Ran 1d K-means. Distance={}".format(distances_by_k[-1]))
# plot_data_opt_k3(distances_by_k, list(range(1,K+1)))
Utils.write_output_csv(cluster, "output.csv", movie_ids)
else:
assert Error("init parameter was not inputted correctly!")
def main():
if len(sys.argv) < 3:
assert Error("need input argument.")
_, csv_path, K = sys.argv
K = int(K)
X = pd.read_csv(csv_path).values
X = PCAHelper.parse_data(X) # Steps 1-5
# k-means and k-means++ execution
km_clusters, km_centroids, km_distances = KMeans.execute(X, K)
kmpp_clusters, kmpp_centroids, kmpp_distances = KMeanspp.execute(X, K)
print("km dist={}, kmpp dist={}".format(km_distances[-1],
kmpp_distances[-1]))
# pca
X = PCAHelper.pca_helper(X, 2)
# plot
Utils.plot_data2(X,
K,
km_clusters,
title="K-means clustering with PCA",
xaxis="First Principal Component",
yaxis="Second Principal Component")
Utils.plot_data2(X,
K,
kmpp_clusters,
title="K-means++ clustering with PCA",
xaxis="First Principal Component",
yaxis="Second Principal Component")
def find_optimum_k():
print("Start find_optimum_k")
iter_per_K = 5
start_range, end_range, step = 1, 40, 2
km_distance_by_K, kmpp_distance_by_K = np.array([]), np.array([])
# Process Input.
X = process_input(sys.argv)
# Try K=1,3,5 .... 29
K_range = range(start_range, end_range, step)
print("Will try K = ", list(K_range))
for K in K_range:
print("Trying K = ", K)
# Run k-means and k-means++ 10 times. Take the average of distance. Append to distance_by_K.
km_dist, kmpp_dist = np.array([]), np.array([])
for _ in range(iter_per_K):
km_clusters, km_centroids, km_distances = KMeans.execute(X, K)
km_clusters, km_centroids, kmpp_distances = KMeanspp.execute(X, K)
km_dist = np.append(km_dist, km_distances[-1])
kmpp_dist = np.append(kmpp_dist, kmpp_distances[-1])
km_ave_dist, kmpp_ave_dist = np.mean(km_dist), np.mean(kmpp_dist)
print("Average Distance | K={} | km={}, kmpp = {}".format(
K, km_ave_dist, kmpp_ave_dist))
km_distance_by_K, kmpp_distance_by_K = np.append(
km_distance_by_K, km_ave_dist), np.append(kmpp_distance_by_K,
kmpp_ave_dist)
print("Plotting Graph")
# 2. Plot graph for K for km_distance_by_K, kmpp_distance_by_K to pick best K.
plot_data_opt_k(km_distance_by_K, kmpp_distance_by_K, list(K_range))
# plot_data_opt_k2(km_distance_by_K, list(K_range), "K-means")
# plot_data_opt_k2(kmpp_distance_by_K, list(K_range), "K-means++")
print("Results:")
print("K range")
print(list(K_range))
print("K-Means Distances:")
print(km_distance_by_K)
print("K-Means++ Distances:")
print(kmpp_distance_by_K)
