update perf plotting

plot/bar.py

@@ -7,18 +7,21 @@
 from plot import ResultData
 
 col0 = [240 / 255, 249 / 255, 232 / 255, 1]
-col1 = [186 / 255, 228 / 255, 188 / 255, 1]
-col2 = [123 / 255, 204 / 255, 196 / 255, 1]
-col3 = [43 / 255, 140 / 255, 190 / 255, 1]
-col4 = [220 / 255, 220 / 255, 220 / 255, 1]
+col1 = [255 / 255, 201 / 255, 71 / 255, 1]
+col2 = [100 / 255, 204 / 255, 197 / 255, 1]
+col3 = [55 / 255, 71 / 255, 79 / 255,
+        1]  # [34 / 255, 40 / 255, 49 / 255, 1]
+col4 = [7 / 255, 102 / 255, 173 / 255, 1]
 col5 = [8 / 255, 104 / 255, 172 / 255, 1]
 light_blue = [166 / 255, 206 / 255, 227 / 255, 1]
 dark_blue = [15 / 255, 90 / 255, 160 / 255, 1]
-white = [1, 1, 1, 1]
+white = [1] * 4
 
 # global options
-plt.rcParams["font.family"] = "Arial"
-plt.rcParams['hatch.linewidth'] = .4
+overall = 'overall'
+patterns = [None, None, None, None, None]
+plt.rcParams['font.family'] = ['Arial']
+plt.rcParams['hatch.linewidth'] = .5
 
 
 def gradient(light, dark):
@@ -41,34 +44,37 @@ def multi_bar(ax, results, cat_names, colors):
     uniq_rl = [x for k, v in groupby(rlbl)
                for x in [k] + [' '] * (sum(1 for __ in v) - 1)]
     labels = [i for i, _ in enumerate(results)]
-    has_ov = llbl[-1].lower() == "overall"
+    has_ov = llbl[-1] == overall
     data = np.array([v for _, v in results])
     data_cum = data.cumsum(axis=1)
     ay = ax.secondary_yaxis('right')
     ax.invert_yaxis()
 
     barh = ([0.6] * (len(llbl) - 2)) + [.8 if has_ov else 0.6]
-    totals = [sum(d) - .75 for d in data[1:]]
+    totals = [sum(d) - .3 for d in data[1:]]
     for i, (name, color) in enumerate(zip(cat_names, colors)):
         widths = data[1:, i]
         starts = data_cum[1:, i] - widths
         ax.barh(labels[1:], widths, left=starts, height=barh,
-                label=name, color=color, zorder=0)
+                label=name, color=color, zorder=0, lw=0,
+                hatch=patterns[i],
+                edgecolor=[.95, .95, .95, 1])
     bars = ax.barh(
-        labels[1:], totals, left=.35, height=barh, color='none',
-        lw=.5, edgecolor=[0, 0, 0, .85], zorder=2, hatch=None)
+        labels[1:], totals, left=.1, height=barh, color='none',
+        lw=0.3, edgecolor=[0, 0, 0, 1], zorder=2, hatch=None)
     for idx in [i for i, tot in enumerate(totals) if tot < 1]:
         bars[idx].set_linewidth(0)
 
-    ax.set_yticklabels(llbl, weight='light')
+    ax.set_yticklabels(llbl)
     ax.set_yticks(np.arange(0, len(llbl), 1))
     ax.set_xlim(0, np.sum(data, axis=1).max())
     ax.yaxis.set_tick_params(length=0)
-    ay.set_yticklabels(uniq_rl, weight='light')
+    ay.set_yticklabels(uniq_rl)
     ay.set_ticks(np.arange(0, len(rlbl), 1))
     ay.yaxis.set_tick_params(length=0)
-    if has_ov:
-        ax.get_yticklabels()[len(llbl) - 1].set_fontweight('bold')
+    for idx, lbl in enumerate(llbl):
+        if idx < len(ax.get_yticklabels()) and lbl == overall:
+            ax.get_yticklabels()[idx].set_fontweight('bold')
     ax.get_yticklabels()[0].set_ha("left")
     ax.get_yticklabels()[0].set_position((.035, 0))
     ay.spines["right"].set_visible(False)
@@ -81,16 +87,17 @@ def multi_bar(ax, results, cat_names, colors):
     ax.set_xticklabels([])
 
 
-def plot_acc(data, mean_data, subplots, plot_name, data_labels,
-             sort_key=None, colors=None, dlen=4):
+def plot_acc(input_data, plot_name, data_labels,
+             sort_key=None, colors=None, dlen=4, overall_bar=True):
+    data, mean_data, subplots = input_data[0]
     if colors is None:
         color_count = len(mean_data)
         colors = get_color_scheme(color_count)
         colors.reverse()
 
     # determine plot size
-    sp_n = len(subplots)
-    plot_height = 1 + 4.5 * (len(data) / 24.)
+    sp_n, pl_n = len(subplots), len(input_data)
+    plot_height = (1 if overall_bar else .7) + 4.5 * (len(data) / 24.)
     h_ratios = [1] if sp_n == 1 else \
         [len([x for x in data.values() if c == x[0][0]]) +
          (3 if i == sp_n - 1 else 0)
@@ -100,30 +107,33 @@ def plot_acc(data, mean_data, subplots, plot_name, data_labels,
 
     # setup figure
     fig, axes = plt.subplots(
-        sp_n, 1, figsize=(3, plot_height),
+        sp_n, pl_n, figsize=(3 * pl_n, plot_height),
         gridspec_kw={'height_ratios': h_ratios})
     ax = axes if sp_n == 1 else axes[-1]
     plt.subplots_adjust(wspace=0, hspace=0)
 
     # draw sub plots
     for i, ckey in enumerate(subplots):
         cdata = [(x[0][1:], x[1]) for x in data.values()
-                 if ckey in x[0][0]]
+                 if ckey == x[0][0]]
         if sort_key is not None:
             cdata.sort(key=sort_key)
         cdata.insert(0, (('', ckey), [0] * dlen))
-        if ckey == subplots[-1]:
+        if ckey == subplots[-1] and overall_bar:
             empty = [((' ', ' '), [0] * dlen)]
-            overall = [v / sum(mean_data) * 100 for v in mean_data]
-            cdata = cdata + empty + [(('', 'overall'), overall)]
+            ov = [v / sum(mean_data) * 100 for v in mean_data]
+            cdata = cdata + empty + [(('', overall), ov)]
         multi_bar(axes if len(subplots) == 1 else axes[i],
                   cdata, data_labels, colors=colors)
 
     # full figure formatting
     leg = fig.legend(data_labels, ncol=2, bbox_to_anchor=(0.22, 1.05),
                      loc='upper left', frameon=False,
-                     handlelength=.9, handletextpad=0.2,
+                     handlelength=.9, handletextpad=0.4,
                      columnspacing=.8, borderpad=0)
+    for p in leg.get_patches():
+        p.set_edgecolor([0, 0, 0, .85])
+        p.set_linewidth(.75)
 
     ax.yaxis.set_tick_params(length=0)
     ax.set_xticks([0, 25, 50, 75, 100])
@@ -133,7 +143,8 @@ def plot_acc(data, mean_data, subplots, plot_name, data_labels,
     ax.xaxis.set_visible(True)
     for i, lbl in enumerate(subplots):
         ax_ = axes if sp_n == 1 else axes[i]
-        shift = 0 if sp_n > 1 and lbl == subplots[-1] else .03
+        shift = 0 if sp_n > 1 and lbl == subplots[-1] else \
+            (.1 if sp_n == 1 else .03)
         box = ax_.get_position()
         box.y0 += shift
         box.y1 += shift
@@ -147,10 +158,18 @@ def plot_acc(data, mean_data, subplots, plot_name, data_labels,
 
 class BarData(ResultData):
 
-    def get_acc_data(self):
-        nums = [BarData.fmt(r) for r in self.raw_rata]
+    def get_acc_data(self, key_test):
+        is_perf = 'perf' in self.directory
+        nums = [BarData.fmt(r) for r in self.raw_rata if key_test(r)]
         means = np.rint(np.mean(np.array(
             [v for _, v in nums]), axis=0)).tolist()
+        if not is_perf:
+            cnums = [BarData.fmt(
+                r, key="Comparison", att="CPGD")
+                for r in self.raw_rata if
+                ResultData.attack(r) == 'CPGD']
+            if len(cnums):
+                nums += cnums
         cats = sorted(list(set([x for ((x, _, _), _) in nums])))
         ndict = dict(enumerate(nums))
         return ndict, means, cats
@@ -165,8 +184,9 @@ def attack(r):
         return ResultData.attack(r).upper()
 
     @staticmethod
-    def fmt(r):
-        keys = (BarData.cls(r), BarData.name(r), BarData.attack(r))
+    def fmt(r, key=None, att=None):
+        keys = (key or BarData.cls(r), BarData.name(r),
+                att or BarData.attack(r))
         valid = round(BarData.valid(r))
         evades = round(BarData.evades(r)) - valid
         accurate = round(BarData.acc(r)) - evades - valid
@@ -177,15 +197,31 @@ def plot_name(self, pattern, out_dir):
         return self.fn_pattern('pdf', pattern, out_dir)
 
 
+def make_plot(bdata, out_dir, is_perf, plot_name, key_test,
+              overall_bar=True):
+    bar_inputs = [[d.get_acc_data(key_test) for d in bdata][0]]
+    # assert (bars have same shape)
+    legend = ['valid', 'evasive', 'accurate', 'inaccurate']
+    plot_acc(bar_inputs,
+             data_labels=legend, overall_bar=overall_bar,
+             plot_name=bdata[0].plot_name(plot_name, out_dir),
+             sort_key=(lambda x: (x[0][0], len(x[0][1]), x[0][1]))
+             if is_perf else
+             (lambda x: (x[0][0], x[0][1])))
+
+
 def plot_bars(data_dir, out_dir=None):
-    bdata = BarData(data_dir)
-    if bdata.n_results:
-        nums, means, cats = bdata.get_acc_data()
-        plot_acc(nums, means, subplots=cats,
-                 data_labels=[
-                     'valid', 'evasive', 'accurate', 'inaccurate'],
-                 plot_name=bdata.plot_name('bar_acc', out_dir),
-                 sort_key=
-                 (lambda x: (x[0][0], x[0][1]))
-                 if 'perf' not in data_dir else
-                 (lambda x: (x[0][0], x[0][1])))
+    dirs = data_dir.split(',')
+    is_perf = 'perf/' in data_dir
+    bdata = [BarData(d) for d in dirs]
+    if is_perf:
+        for cats in ['PT-1', 'PT-2']:
+            key_test = lambda r: cats in BarData.name(r)
+            if len(bdata) and bdata[0].n_results:
+                make_plot(
+                    bdata, out_dir, is_perf, 'bar_acc_' + cats,
+                    key_test, overall_bar=False)
+    else:
+        key_test = lambda r: ResultData.attack(r) != 'CPGD'
+        if len(bdata) and bdata[0].n_results:
+            make_plot(bdata, out_dir, is_perf, 'bar_acc', key_test)