add script for plots

Signed-off-by: Lagrang3 <[email protected]>

Author: Lagrang3

mcf/algorithm.c

@@ -1036,7 +1036,7 @@ bool solve_constrained_fcnfp(const tal_t *ctx, const struct graph *graph,
 	 * coded maximum iterations. */
 	const size_t FCNFP_iterations = 10;
 	const size_t first_round_FCNFP_iterations = 100;
-	const double decay_exponent = 0.9;
+	const double decay_exponent = 0.5;
 
 	const size_t max_num_arcs = graph_max_num_arcs(graph);
 	const size_t max_num_nodes = graph_max_num_nodes(graph);

python/fuzzy-constrained-fcnfp.py

@@ -149,7 +149,8 @@ def generate_problem(
     attempts = 0
     while True:
         capacity = np.random.randint(1, M_cap + 1, size=N_arcs)
-        cost = np.random.randint(M_cost + 1, size=(N_constraints, N_arcs))
+        # cost is never zero
+        cost = np.random.randint(1, M_cost + 1, size=(N_constraints, N_arcs))
         fixedcost = np.random.randint(M_charge + 1, size=(N_constraints, N_arcs))
 
         # find out the max flow

scatter-plot-cfcnfp.py

@@ -0,0 +1,104 @@
+from sys import stdin
+import math
+import numpy as np
+import matplotlib.pyplot as plt
+
+n_constraints = 0
+fixed_charge = False
+
+def transform_scale(x):
+    return -np.log2(1.0 - x)
+
+def mkboxes(X, Y):
+    pos = list(set([i for i in X]))
+    pos.sort()
+    idx = {}
+    for i, x in enumerate(pos):
+        idx[x] = i
+    boxes = [ [] for i in range(len(pos))]
+    for x,y in zip(X,Y):
+        boxes[idx[x]].append(y)
+    return boxes, pos
+
+def plot_scatter(X, Y):
+    quantiles = [0.5, 0.75, 0.95, 0.99]
+
+    X_transf = transform_scale(X)
+    boxes, positions = mkboxes(X_transf, Y)
+    X_ticks = [i for i in range(2+int(max(X_transf)))]
+    X_labels = ["%.2f" % (1. - 1./2**i) for i in X_ticks]
+    
+    fig, ax = plt.subplots()
+    ax.set_ylim(0.999, 1.6)
+    # ax.boxplot(boxes, positions=positions)
+    ax.violinplot(boxes, positions=positions)
+    ax.set_xticks(X_ticks)
+    ax.set_xticklabels(X_labels)
+    ax.set_xlabel("difficulty")
+    ax.set_ylabel("approximate/best solution ratio")
+    ax.set_title("%d-constraints solution accuracy (%s activation costs)" % (n_constraints-1,
+        "with" if fixed_charge else "without"))
+    
+    qvalues = np.array([ np.quantile(b, quantiles) for b in boxes ])
+    qvalues = qvalues.transpose()
+    
+    for i in range(len(quantiles)):
+        ax.plot(positions, qvalues[i], label = "%.2f%%" % (100*quantiles[i]))
+    
+    plt.legend()
+    plt.show()
+
+
+def plot_successrate(X, Y):
+    X_transf = transform_scale(X)
+    boxes, positions = mkboxes(X_transf, Y)
+    
+    positions = [float(p) for p in positions ]
+    Y_rate = [ float(sum(b)/len(b)) for b in boxes ]
+    
+    X_ticks = [i for i in range(2+int(max(X_transf)))]
+    X_labels = ["%.2f" % (1. - 1./2**i) for i in X_ticks]
+    
+    fig, ax = plt.subplots()
+    ax.plot(positions, Y_rate, ls='', marker='o')
+    
+    ax.set_xticks(X_ticks)
+    ax.set_xticklabels(X_labels)
+    ax.set_xlabel("difficulty")
+    ax.set_ylabel("success rate")
+    ax.set_title("%d-constraints success rate (%s activation costs)" % (n_constraints-1, 
+        "with" if fixed_charge else "without"))
+    plt.show()
+
+# all results
+DIFF = []
+SUCCESS = []
+
+# only those results where constraints are satisfied
+DIFF_filtered = []
+COST_relative = []
+
+for line in stdin:
+    elements = line.split()
+    pid = elements[0]
+    d = float(elements[1])
+    exact, nc, mcf, nc_mcf, approx, nc_approx = map(int, elements[2:])
+    
+    n_constraints = nc
+    
+    DIFF.append(d)
+    # constraints satisfied
+    assert nc_approx<=nc
+    SUCCESS.append(int(nc_approx==nc))
+    
+    if nc_approx==nc:
+        DIFF_filtered.append(d)
+        # the approximate is never better than the exact
+        assert exact<=approx
+        COST_relative.append(approx/exact)
+        
+        if COST_relative[-1]>1.5:
+            print(line)
+
+plot_scatter(np.array(DIFF_filtered), np.array(COST_relative))
+plot_successrate(np.array(DIFF), np.array(SUCCESS))