update k_confidence test

.gitignore

@@ -5,4 +5,5 @@ __pycache__/
 *$py.class
 *.pkl
 temp*.py
-nas/config.json
+nas/config.json
+record.txt

main.py

@@ -249,11 +249,11 @@ def parse_value(_val):
 
 def main_benchmark(problem_name: str):
     _seeds = 20
-    _n_proc = 1
+    _n_proc = 20
     init_seed = 42
-    _estimate_gram = 2.6
+    _estimate_gram = 3.5
     usage_check(_n_proc)
-    gpu_ids = [0]
+    gpu_ids = [0, 1, 2, 3, 4, 5, 6, 7]
     _res = benchmark_for_seeds(main,
                                post_mean_std,
                                seeds=_seeds,
@@ -300,4 +300,4 @@ def fast_seed(seed: int) -> None:
         'd7': 'DTLZ7b',
     })
     # main(problems.d7, do_plot=True, print_progress=True, do_train=True)
-    main_benchmark(problems.d7)
+    main_benchmark(problems.d1)

problem_config/example.py

@@ -13,9 +13,9 @@ def get_args():
     args = NamedDict()
     args.problem_dim = (10, 3)
     args.train_test = (300, 1)
-    args.epoch = 50
-    args.sgd_epoch = 50
-    args.sgd_select_n = 50
+    args.epoch = 10
+    args.sgd_epoch = 10
+    args.sgd_select_n = 100
     args.update_lr = 0.0025
     args.meta_lr = 0.001
     args.fine_tune_lr = 0.005

utils.py

@@ -32,10 +32,10 @@ def set_ipython_exception_hook():
     sys.excepthook = _IPythonExceptionHook()
 
 
-def test():
+def test(_n, _m):
     import numpy as np
-    tot = 900
-    each = 50
+    tot = _n
+    each = _m
     tot_arr = np.arange(tot, dtype=np.int32)
     sel = np.zeros_like(tot_arr, dtype=bool)
     i = 0
@@ -84,7 +84,7 @@ def draw_curve(n, m):
     import matplotlib.pyplot as plt
     k = list(range(10, 200))
     y = [((n ** ki - (n - m) ** ki) / n ** ki) ** n for ki in k]
-    arr = [test() for _ in range(2000)]
+    arr = [test(n, m) for _ in range(2000)]
     plt.hist(arr, bins=50, density=True)
     div = np.diff(y) / np.diff(k)
     div = np.concatenate(([div[0]], div))
@@ -94,7 +94,7 @@ def draw_curve(n, m):
 
 
 if __name__ == '__main__':
-    _n, _m = 900, 50
-    draw_curve(_n, _m)
+    _n, _m = 900, 100
+    # draw_curve(_n, _m)
     v = calculate_confidence_k(_n, _m)
     print(v)