Add comment: Change parameters for MNIST dataset

README.md

@@ -6,10 +6,11 @@ Beyond", published in ICML 2022.
 ## Additional Test that we excecuted
 1. **Less-separated Synthetic Dataset**: run
 `python experiments.py complete`
-- 
+- Change 'r' value at https://github.com/dom-lee/EECS553-Reproducibility-Spectral-Clustering/blob/06541f4ec59481cefdc981144c0b80f75715a451/pysc/datasets.py#L379
 
 2. **Test on BSDS dataset with different standard deviation**: run
 `python experiments.py bsds`
+- We set break condition to cluster only 25 images
 - 
 
 3. **Test on MNIST dataset with different number of eigenvector for embedding**:

experiments.py

@@ -88,23 +88,26 @@ def experiment_instance(d, nn, q):
             pysc.datasets.MnistDataset(k=nn, downsample=d), 10)
         q.put((d, nn, this_rand_scores, this_mut_info, this_conductances, this_times))
 
-    # We will construct the 3-NN graph for the MNIST dataset.
-    k = 3
-
-    # Kick off the experiment in a sub-process.
-    q = Queue()
-    p = Process(target=experiment_instance, args=(None, k, q))
-    p.start()
-    p.join()
+    # # We will construct the 3-NN graph for the MNIST dataset.
+    # k = 3
+
+    # [Hyungchan Cho] We changed k parameter from 3 to 21 to generate Graph with KNN
+    for k in range(3, 21):
+        # Kick off the experiment in a sub-process.
+        q = Queue()
+        p = Process(target=experiment_instance, args=(None, k, q))
+        p.start()
+        p.join()
 
-    # Write out the results
-    with open("results/mnist/results.csv", 'w') as fout:
-        fout.write("k, d, eigenvectors, rand\n")
+        # Write out the results
+        with open(f"results/mnist/results_k.csv", 'w') as fout:
+            fout.write("k, d, eigenvectors, rand\n")
 
-        while not q.empty():
-            downsample, k, rand_scores, _, _, _ = q.get()
-            for i in range(2, 11):
-                fout.write(f"{k}, {downsample}, {i}, {rand_scores[i]}\n")
+            while not q.empty():
+                downsample, k, rand_scores, _, _, _ = q.get()
+                # [Hyungchan Cho] Tested k eigenvectors from 2 to 21
+                for i in range(2, 21):
+                    fout.write(f"{k}, {downsample}, {i}, {rand_scores[i]}\n")
 
 
 def run_usps_experiment():

results/sbm/complete_results.csv

@@ -1 +1,6 @@
 k, n, p, q, poverq, eigenvectors, conductance, rand
+5, 1000, 0.01, 0.001, 10.0, 1, 0.9474027622503941, 0.6574567073414683
+5, 1000, 0.01, 0.001, 10.0, 2, 0.6436893037077096, 0.8146371834366872
+5, 1000, 0.01, 0.001, 10.0, 3, 0.3756434337075715, 0.9512978275655131
+5, 1000, 0.01, 0.001, 10.0, 4, 0.31325660254706367, 0.9812917143428687
+5, 1000, 0.01, 0.001, 10.0, 5, 0.28187226965267426, 0.995807473494699