def fade(widget, smoothness=3, cnf={}, **kw):
"""This function will show faded effect on widget's different color options.
Args:
widget (tk.Widget): Passed by the bind function.
smoothness (int): Set the smoothness of the fading (1-10).
background (str): Fade background color to.
foreground (str): Fade foreground color to."""
kw = tk._cnfmerge((cnf, kw))
if not kw: raise ValueError("No option given, -bg, -fg, etc")
if len(kw) > 1:
return [fade(widget, smoothness, {k: v}) for k, v in kw.items()][0]
if not getattr(widget, '_after_ids', None): widget._after_ids = {}
widget.after_cancel(widget._after_ids.get(list(kw)[0], ' '))
c1 = tuple(map(lambda a: a / 65535, widget.winfo_rgb(widget[list(kw)[0]])))
c2 = tuple(map(lambda a: a / 65535,
widget.winfo_rgb(list(kw.values())[0])))
colors = tuple(
colour.rgb2hex(c, force_long=True)
for c in colour.color_scale(c1, c2, max(1, smoothness * 15)))
def worker(count=0):
if len(colors) - 1 <= count: return
widget.config({list(kw)[0]: colors[count]})
widget._after_ids.update({
list(kw)[0]:
widget.after(max(1, int(smoothness / 10)), worker, count + 1)
})
worker()
def set_color_range(self, color1, color2):
hsl_cs = colour.color_scale(color1.get_hsl(), color2.get_hsl(), 100)
rgb_cs = [colour.hsl2rgb(a) for a in hsl_cs]
self.scaled_cs = []
for a in rgb_cs:
self.scaled_cs.append(
(round(a[0] * 255), round(a[1] * 255), round(a[2] * 255)))
def dataset_viz(mu_df, targets):
"""Creates histograms for given dataset, type filter and targets
Args:
mu_df (pd.DataFrame): A dataframe holding all failures
targets (dict(str,list(str))): Column names from mu_df to perform analysis on
"""
def lengths(x):
if isinstance(x, list):
yield len(x)
for y in x:
yield from lengths(y)
rows = len(targets.keys())
cols = max(lengths(list(targets.values())))
fig, axes_list = plt.subplots(rows, cols)
axes_list[-1, -1].axis('off')
fig.set_size_inches(18, 8)
meta_c_df = pd.read_csv('datasets/study_classification_info.csv')
meta_r_df = pd.read_csv('datasets/study_regression_info.csv')
meta_df = pd.concat([meta_c_df, meta_r_df])
row_size = max([len(x) for x in targets.values()])
base_colors = [
hsl2hex(c)
for c in color_scale((0., 0.8, 0.6), (0.8, 0.8, 0.6), row_size)
]
for j, TYPE in enumerate(targets):
for i, BASE in enumerate(targets[TYPE]):
all_data = pd.merge(mu_df.loc[mu_df['TYPE'] == TYPE],
meta_df,
how='left')
full_data = pd.merge(mu_df, meta_df, how='left')
ax = axes_list[j][i]
ax.set_xlabel(BASE.capitalize() + " Count (Log Scale)")
ax.set_ylabel("{} Frequency".format(BASE.capitalize()))
counts, bins, bars = ax.hist(all_data[BASE],
bins=np.logspace(
np.log10(np.min(full_data[BASE])),
np.log10(np.max(full_data[BASE])),
30),
stacked=True,
color=base_colors[i],
alpha=0.7,
edgecolor='black',
linewidth=0.6)
ax.set_xscale('log')
fig.suptitle('Content Analysis of Datasets')
fig.subplots_adjust(hspace=0.4, wspace=0.3)
if not os.path.exists('figures'):
os.makedirs('figures')
plt.savefig('figures/DatasetShapes.pdf', dpi=fig.dpi, transparent=True)
def color_fade(widget, **kw):
if not getattr(widget, '_after_ids', None):
widget._after_ids = {}
widget.after_cancel(widget._after_ids.get(list(kw)[0], ' '))
color_a = tuple(c / 65535 for c in widget.winfo_rgb(widget[list(kw)[0]]))
color_b = tuple(c / 65535 for c in widget.winfo_rgb(list(kw.values())[0]))
colors = tuple(
colour.rgb2hex(color, force_long=True)
for color in colour.color_scale(color_a, color_b, 70))
def update_widget_after(count=0):
if len(colors) - 1 <= count:
return
else:
widget.config({list(kw)[0]: colors[count]})
widget._after_ids.update(
{list(kw)[0]: widget.after(1, update_widget_after, count + 1)})
update_widget_after()
def boxplot_viz(clean_df, target):
clean_df = clean_df[target]
models = pd.unique(clean_df.index.values)
data_arr = np.array([clean_df[m].values for m in models]).T
base_colors = [
hsl2hex(c)
for c in color_scale((0., 0.8, 0.6), (0.8, 0.8, 0.6), len(models))
]
plt.figure(figsize=(7, 3.5))
title_str = "Raw Per Model {} Comparison ({})".format(
'Classification' if target == 'F1_SCORE' else 'Regression', target)
plt.title(title_str, size=12)
bplot = plt.boxplot(data_arr,
vert=False,
patch_artist=True,
notch=True,
labels=" ",
positions=list(reversed(range(1,
len(models) + 1))))
for p, c in zip(bplot['boxes'], base_colors):
p.set_facecolor(c)
plt.legend(bplot['boxes'],
models,
loc='lower left',
prop={'size': 8},
fancybox=True,
framealpha=0.6)
plt.setp(bplot['fliers'], markeredgecolor='grey')
plt.setp(bplot['medians'], color='black')
# plt.show()
plt.savefig('figures/RawDataBoxPlot{}.pdf'.format(target),
dpi=plt.gcf().dpi,
transparent=True)
def rg_gradient(start, end, point):
gradient = colour.color_scale(colour.web2hsl('red'),
colour.web2hsl('green'), 20)
stepsize = (end - start) / 20
gradient_bin = np.clip(0, 19, int((point - start) / stepsize))
return colour.hsl2web(gradient[gradient_bin])
def rg_gradient(start, end, point):
gradient = colour.color_scale(colour.web2hsl(
'red'), colour.web2hsl('green'), 20)
stepsize = (end - start) / 20
gradient_bin = np.clip(0, 19, int((point - start) / stepsize))
return colour.hsl2web(gradient[gradient_bin])
def pairwise_comp_viz(mu_df, target):
"""Creates a pariwise interaction visualization plot comparing each model against the other
Args:
mu_df (pd.Dataframe): A dataframe of valid runs with type and model as indicies with aggregated
means across runs
c_df_info (pd.Dataframe): A dataframe with information about each dataset type
"""
def plot_comp(mu_df, m1, m2, target, vmin, vmax, cmap, ax):
m1_values = mu_df.xs(m1, level=1).values
m2_values = mu_df.xs(m2, level=1).values
# difference from y=x color mapping (not magnitude because independent)
colors = np.array(
[m_2 - m_1 for m_2, m_1 in zip(m2_values, m1_values)])
sc = ax.scatter(m1_values,
m2_values,
alpha=0.7,
s=15,
c=colors,
cmap=cmap,
zorder=10,
norm=MidpointNormalize(vmin=vmin,
vmax=vmax,
midpoint=0))
ax.set_xlabel(m1)
ax.set_ylabel(m2)
lims = [
np.min([ax.get_xlim(), ax.get_ylim()]),
np.max([ax.get_xlim(), ax.get_ylim()])
]
ax.plot(lims, lims, 'k-', lw=0.7, alpha=0.7, zorder=0)
ax.set_aspect('equal')
ax.set_xlim(lims)
ax.set_ylim(lims)
return sc
class MidpointNormalize(mpl.colors.Normalize):
def __init__(self, vmin=None, vmax=None, midpoint=None, clip=False):
self.midpoint = midpoint
mpl.colors.Normalize.__init__(self, vmin, vmax, clip)
def __call__(self, value, clip=None):
x, y = [self.vmin, self.midpoint, self.vmax], [0, 0.5, 1]
return np.ma.masked_array(np.interp(value, x, y))
def get_color_range(c1, c2, bins):
MAX_L = 0.7
c1h, c1s, c1l = c1.hsl
c2h, c2s, c2l = c2.hsl
c1_bins = [
Color(hsl=(c1h, c1s, var_l))
for var_l in np.linspace(c1l, MAX_L, int(bins / 2))
]
c2_bins = [
Color(hsl=(c2h, c2s, var_l))
for var_l in np.linspace(MAX_L, c2l, int(bins / 2))
]
color_range = [c.hex_l for c in (c1_bins + c2_bins)]
return color_range
sort_order = {'auto-sklearn': 1, 'tpot': 2, 'h2o': 3, 'auto_ml': 4}
mu_df = mu_df[target]
models = sorted(pd.unique(mu_df.index.get_level_values('MODEL').values),
key=lambda x: sort_order[x])
combos = list(itertools.combinations(models, 2))
sorted_combos = list(
sorted(combos, key=lambda x: (sort_order[x[0]], sort_order[x[1]])))
plot_count = len(sorted_combos)
rows, cols = square_fac(plot_count)
fig, ax_list = plt.subplots(rows, cols)
fig.set_size_inches(17, 8)
metric_name = target.replace(
'_', ' ').title() if target == 'F1_SCORE' else target
base_colors = [
hsl2hex(c)
for c in color_scale((0, 0.7, 0.4), (1, 0.7, 0.4), plot_count)
]
model_colors = {m: c for m, c in zip(models, base_colors)}
color_bins = 10
scatters = []
# get min-max of differences
vmin = np.inf
vmax = -np.inf
for m1, m2 in sorted_combos:
m1_values = mu_df.xs(m1, level=1).values
m2_values = mu_df.xs(m2, level=1).values
colors = np.array(
[m_2 - m_1 for m_2, m_1 in zip(m2_values, m1_values)])
if np.max(colors) > vmax:
vmax = np.max(colors)
if np.min(colors) < vmin:
vmin = np.min(colors)
for combo, ax in zip(sorted_combos, ax_list.ravel()):
m1, m2 = combo
color_range = get_color_range(Color(model_colors[m1]),
Color(model_colors[m2]), color_bins)
cmap = mpl.colors.ListedColormap(color_range)
scatters.append(plot_comp(mu_df, m1, m2, target, vmin, vmax, cmap, ax))
for sc, ax in zip(scatters, ax_list.ravel()):
cbar = fig.colorbar(sc, ax=ax, fraction=0.046, pad=0.08)
cbar.ax.tick_params(labelsize=10)
ax_str = '{} Difference' if target == 'F1_score' else 'Standardized Inverted {} Difference'
cbar.set_label(ax_str.format(metric_name),
rotation=90,
fontsize=8,
labelpad=-57) # if target == 'F1_SCORE' else -65)
fig.suptitle('Dataset Mean {} Across Frameworks'.format(metric_name))
if not os.path.exists('figures'):
os.makedirs('figures')
plt.savefig('figures/DatasetMean{}.pdf'.format(metric_name.replace(
' ', '')),
dpi=fig.dpi,
transparent=True)
def correlation_viz(mu_df, targets):
"""Creates scatterplots of correlation betwene dataset stats and model performance
Args:
mu_df (pd.DataFrame): A dataframe holding all failures
targets (dict(str,list(str))): Column names from mu_df to perform analysis on
"""
def get_true_features(d_id):
df = pd.read_csv('datasets/{}.csv'.format(d_id))
df_types = pd.read_csv('datasets/{}_types.csv'.format(d_id))
#Get categorical encoded column count
df_types_cat = df_types.loc[df_types['TYPE'] == 'categorical']['NAME']
df_cat = df[df.columns.intersection(df_types_cat.values)]
uniques = [len(df_cat[col].unique()) for col in df_cat]
df_types_num = df_types.loc[df_types['TYPE'] == 'numerical']['NAME']
df_num = df[df.columns.intersection(df_types_num.values)]
count = np.sum(uniques) + len(df_num.columns)
return count
plt.gcf().set_size_inches(20, 15)
meta_c_df = pd.read_csv('datasets/study_classification_info.csv')
meta_r_df = pd.read_csv('datasets/study_regression_info.csv')
meta_df = pd.concat([meta_c_df, meta_r_df])
meta_df['DIMENSIONALITY'] = meta_df.apply(
lambda row: get_true_features(row['DATASET_ID']), axis=1)
full_data = pd.merge(mu_df, meta_df, how='left')
models = full_data['MODEL'].unique()
row_size = max([len(x) for x in targets.values()])
base_colors = [
hsl2hex(c)
for c in color_scale((0., 0.8, 0.6), (0.8, 0.8, 0.6), len(models))
]
lines = None
for j, TYPE in enumerate(targets):
for i, BASE in enumerate(targets[TYPE][1]):
all_data = full_data.loc[full_data['TYPE'] == TYPE]
all_data = all_data[[
'MODEL', 'DATASET_ID', BASE, targets[TYPE][0]
]]
all_data = all_data.groupby(['MODEL', 'DATASET_ID', BASE],
as_index=False).mean()
all_data = all_data.sort_values(BASE)
plt.subplot(len(targets), row_size, row_size * j + i + 1)
ylabel_str = targets[TYPE][0]
if ylabel_str.lower() == 'mse':
label_str = 'standardized negated mse'
plt.xlabel(BASE.replace('_', ' ').capitalize())
plt.ylabel("{} {}".format(
TYPE.replace('_', ' ').capitalize(),
ylabel_str.replace('_', ' ').capitalize()))
local_lines = []
for k, m in enumerate(models):
ss = all_data.loc[all_data['MODEL'] == m]
ss[BASE] = ss[BASE].rolling(int(len(ss[BASE]) / 2)).median()
ss[targets[TYPE][0]] = ss[targets[TYPE][0]].rolling(
int(len(ss[BASE]) / 2)).median()
x = ss[BASE]
y = ss[targets[TYPE][0]]
line, = plt.plot(x,
y,
color=base_colors[k],
alpha=0.7,
label=m)
local_lines.append(line)
if lines == None:
lines = local_lines
plt.figlegend(lines, models, fancybox=True, framealpha=0.0)
plt.gcf().suptitle('Dataset Dependent Performance Analysis')
if not os.path.exists('figures'):
os.makedirs('figures')
plt.savefig('figures/DatasetPerformance.pdf',
dpi=plt.gcf().dpi,
transparent=True)
cyanExtruder = 2
magentaExtruder = 0
yellowExtruder = 1
def cambiacolor(color, v_ext=0):
cyan, magenta, yellow = color
result = "M163 S%s P%s\n" % (cyanExtruder, cyan)
result += "M163 S%s P%s\n" % (magentaExtruder, magenta)
result += "M163 S%s P%s\n" % (yellowExtruder, yellow)
result += "M164 S%s\n" % v_ext
result += "T%s" % v_ext
return result
while True:
try:
line = raw_input()
except EOFError:
break
if ";LAYER_COUNT:" in line:
layers = int(line[len(";LAYER_COUNT:"):])
scale = c.color_scale((0, 0, 1),
(0, 1, 0), int(layers / 2)) + c.color_scale(
(0, 1, 0), (1, 0, 0), int(layers / 2))
elif ";LAYER:" in line:
print(line)
print(cambiacolor(scale.pop(0)))
else:
print(line)