def plot_all_csvs_from_multiple_files_split_diff_datasets(csv_paths_files_list, baseline_st_idx, y_axis):
assert len(csv_paths_files_list) == 3, "should have 3 files for 3 bulk experiments"
gs = gridspec.GridSpec(1, 3)
axis_mat = np.empty([5, 3], dtype=object)
fig_arr = []
for i in range(5):
fig = define_new_figure(size=(20, 5))
fig.set_facecolor("#FFFFFF")
fig_arr.append(fig)
for j in range(3):
ax = fig.add_subplot(gs[0, j])
axis_mat[i, j] = ax
for j, bulk_expr_csv_f in enumerate(csv_paths_files_list):
csv_file_paths_list = file_util.readlines(bulk_expr_csv_f)
for f_num, f_path in enumerate(csv_file_paths_list):
x_ax, x_ax_vals, our_ys, our_lbls, baselines_ys, baselines_lbls = extract_info_from_csv(f_path,
baseline_st_idx)
plot_experiment_publish_format(axis_mat[f_num, j], x_ax_vals, our_ys, our_lbls,
baselines_ys, baselines_lbls, x_ax, y_axis)
share_axis_between_rows(axis_mat, axis='y')
#plt.tight_layout()
csv_file_paths_list = file_util.readlines(csv_paths_files_list[0])
datasets_names = map(lambda p: os.path.splitext(os.path.basename(p))[0].split('_')[0], csv_file_paths_list)
for i, fig in enumerate(fig_arr):
p = os.path.dirname(os.path.dirname(csv_paths_files_list[0])) + "/" + datasets_names[i] + ".pdf"
print p
save_figure_publishable(fig, p)
plt.clf()
def plot_experiment_effect_of_num_of_operators_publishable(expr_fold_path):
import csv
fig = define_new_figure(size=(10, 5))
fig.set_facecolor("#FFFFFF")
ax = fig.add_subplot(111)
csv_f_path, all_exprs_data_points = extract_info_from_expr_folder(expr_fold_path)
with open(csv_f_path, 'rb') as f:
reader = csv.reader(f, delimiter='\t')
tbl_headers = reader.next()
x_axis = tbl_headers[0]
our_labels = [['']]
baselines_labels = []
x_axis_vals = []
our_ys = [[]]
baselines_ys = []
for line in reader:
x_axis_vals.append(int(line[0]))
our_ys[0].append(float(line[2]))
plot_experiment_publish_format(ax, x_axis_vals, our_ys, our_labels,
baselines_ys, baselines_labels, x_axis, 'final accuracy')
# print list(csv.reader(open(csv_f_path, 'r'), delimiter='\t'))[1:]
baselines_ys = []
plot_experiment_publish_format(ax, x_axis_vals, our_ys, our_labels,
baselines_ys, baselines_labels, x_axis, "final accuracy")
#plt.tight_layout()
# ax.set_ylim(0.55, 0.70)
save_figure_publishable(fig, os.path.dirname(csv_f_path) + "/op_num_effect.pdf")
plt.show()
def plot_all_csvs_from_file(fold_paths_file, y_axis):
fig = define_new_figure(size=(15, 14))
fig.set_facecolor("#FFFFFF")
gs = gridspec.GridSpec(3 * 2, 4 * 3)
axis_list = []
for i in range(2):
for j in range(2):
ax = fig.add_subplot(gs[i * 2:(i + 1) * 2, j * 2 * 3:(j + 1) * 2 * 3])
axis_list.append(ax)
ax = fig.add_subplot(gs[4:6, 0:6])
axis_list.append(ax)
# fig = define_new_figure(size=(20, 8))
# fig.set_facecolor("#FFFFFF")
#
# gs = gridspec.GridSpec(2, 2 * 3)
# axis_list = []
# for i in range(3):
# for j in range(1):
# ax = fig.add_subplot(gs[0, i * 2:(i + 1) * 2])
# axis_list.append(ax)
# ax = fig.add_subplot(gs[1, 0:2])
# axis_list.append(ax)
# ax = fig.add_subplot(gs[1, 2:4])
# axis_list.append(ax)
titles = ('CMR', 'SUBJ', 'SST', 'HS', 'KS')
fold_paths_list = file_util.readlines(fold_paths_file)
for f_num, f_path in enumerate(fold_paths_list):
csv_path, _ = extract_info_from_expr_folder(f_path)
x_ax, x_ax_vals, our_ys, our_lbls, baselines_ys, baselines_lbls = extract_info_from_csv(csv_path, 4)
our_lbls = ['US-BS-MQ', 'US-HC-MQ', 'S-MQ']
baselines_lbls = ['IDEAL-RAND', 'WNA', 'RNN-LM']
axis_list[f_num].set_title(titles[f_num], fontsize=32)
plot_experiment_publish_format(axis_list[f_num], x_ax_vals, our_ys, our_lbls,
baselines_ys[:-1], baselines_lbls[:-1], x_ax, y_axis)
#plt.tight_layout(pad=0.3)
plt.legend(bbox_to_anchor=(1.3, 0.95), loc=2, prop={'size': 25})
save_figure_publishable(fig, os.path.dirname(fold_paths_file) + "/all_plots_publish.png") # eps the format conferences want
# for some reason the showed plot is different from the one saved
plt.show()
def plot_all_final_results_from_file_with_error_bars(fold_paths_file):
us_hc_scores = []
us_hc_stds = []
us_beam_scores = []
us_beam_stds = []
rand_syn_scores = []
rand_syn_stds = []
wna_scores = []
wna_stds = []
ind = np.arange(5) # the x locations for the groups
width = 0.20 # the width of the bars
fig = define_new_figure(size=(10, 4.5))
fig.set_facecolor("#FFFFFF")
ax = fig.add_subplot(111)
exprs_fold_paths = file_util.readlines(fold_paths_file)
for f_num, fold_path in enumerate(exprs_fold_paths):
csv_path, all_exprs_data_points = extract_info_from_expr_folder(fold_path)
x_axis, x_axis_vals, our_ys, our_labels, baselines_ys, baselines_labels = extract_info_from_csv(csv_path, 4)
ideal_scr = baselines_ys[0][-1]
all_exprs_data_points = map(
lambda all_res: map(lambda data_p_lst: [x / ideal_scr for x in data_p_lst], all_res),
all_exprs_data_points)
all_expr_data_points_std = extract_std_deviation_all_exprs_data_points(all_exprs_data_points)
us_beam_scores.append(our_ys[0][-1] / ideal_scr)
us_beam_stds.append(all_expr_data_points_std[0][-1] / 2)
us_hc_scores.append(our_ys[1][-1] / ideal_scr)
us_hc_stds.append(all_expr_data_points_std[1][-1] / 2)
rand_syn_scores.append(our_ys[2][-1] / ideal_scr)
rand_syn_stds.append(all_expr_data_points_std[2][-1] / 2)
wna_scores.append(baselines_ys[1][-1] / ideal_scr)
wna_stds.append(all_expr_data_points_std[4][-1] / 2)
# init_score = baselines_ys[0][0]
# all_exprs_data_points = map(lambda all_res: map(lambda data_p_lst: [(x-init_score)/(ideal_scr-init_score) for x in data_p_lst], all_res),
# all_exprs_data_points)
# all_expr_data_points_std = extract_std_deviation_all_exprs_data_points(all_exprs_data_points)
#
#
# us_beam_scores.append((our_ys[0][-1]-init_score)/(ideal_scr-init_score))
# us_beam_stds.append(all_expr_data_points_std[0][-1] / 2)
# us_hc_scores.append((our_ys[1][-1]-init_score)/(ideal_scr-init_score))
# us_hc_stds.append(all_expr_data_points_std[1][-1] / 2)
# rand_syn_scores.append((our_ys[2][-1]-init_score)/(ideal_scr-init_score))
# rand_syn_stds.append(all_expr_data_points_std[2][-1] / 2)
# wna_scores.append((baselines_ys[1][-1]-init_score)/(ideal_scr-init_score))
# wna_stds.append(all_expr_data_points_std[4][-1] / 2)
ax.set_xticks(ind)
ax.bar(ind - 1.5 * width, us_hc_scores, width, yerr=us_hc_stds,
color='SkyBlue', label='US-HC-MQ', hatch="-")
ax.bar(ind - 0.5 * width, us_beam_scores, width, yerr=us_beam_stds,
color='IndianRed', label='US-BS-MQ', hatch="\\\\")
ax.bar(ind + 0.5 * width, rand_syn_scores, width, yerr=rand_syn_stds,
color='gold', label='S-MQ', hatch="//")
ax.bar(ind + 1.5 * width, wna_scores, width, yerr=wna_stds,
color='purple', label='WNA')
# minlim, maxlim = ax.get_xlim()
# ax.plot([minlim, maxlim], [0., 0.], "k--")
# ax.set_xlim([minlim, maxlim])
ax.set_ylim(0.5, 1.05)
# Add some text for labels, title and custom x-axis tick labels, etc.
ax.set_ylabel('IDEAL accuracy ratio')
ax.set_xticklabels(('CMR', 'SUBJ', 'SST', 'HS', 'KS'))
#plt.tight_layout()
ax.legend(loc='best')
plt.show()
save_figure_publishable(fig, os.path.dirname(fold_paths_file) + "/final_results_error_bars.png")
def plot_label_switch_difference_publishable(fold_paths_file):
us_hc_scores = []
us_hc_stds = []
rand_hc_scores = []
rand_hc_stds = []
rand_syn_scores = []
rand_syn_stds = []
ind = np.arange(5) # the x locations for the groups
width = 0.25 # the width of the bars
inst_gen_c = 50 # instance gen count
fig = define_new_figure(size=(10, 3.5))
fig.set_facecolor("#FFFFFF")
ax = fig.add_subplot(111)
exprs_fold_paths = file_util.readlines(fold_paths_file)
for f_num, fold_path in enumerate(exprs_fold_paths):
csv_path, all_exprs_data_points = extract_info_from_expr_folder(fold_path)
all_exprs_data_points = map(
lambda all_res: map(lambda data_p_lst: [float(x) / inst_gen_c for x in data_p_lst], all_res),
all_exprs_data_points)
all_expr_data_points_std = extract_std_deviation_all_exprs_data_points(all_exprs_data_points)
with open(csv_path, 'rb') as f:
csv_lines = list(csv.reader(f, delimiter='\t'))
cells = map(lambda a: float(a) / inst_gen_c, csv_lines[1]) # line with semantic env 10
us_hc_scores.append(cells[1])
us_hc_stds.append(all_expr_data_points_std[0][1] / 2)
rand_hc_scores.append(cells[2])
rand_hc_stds.append(all_expr_data_points_std[1][1] / 2)
rand_syn_scores.append(cells[3])
rand_syn_stds.append(all_expr_data_points_std[2][1] / 2)
rects1 = ax.bar(ind - width, us_hc_scores, width, hatch='-', # yerr=us_hc_stds,
color='SkyBlue', label='US-HC-MQ')
rects2 = ax.bar(ind, rand_hc_scores, width, hatch='\\\\', # yerr=rand_hc_stds,
color='IndianRed', label='S-HC-MQ')
rects3 = ax.bar(ind + width, rand_syn_scores, width, hatch='//', # yerr=rand_syn_stds,
color='gold', label='S-MQ')
# Add some text for labels, title and custom x-axis tick labels, etc.
ax.set_ylabel('% of switched instances')
ax.set_xticks(ind)
ax.set_xticklabels(('CMR', 'SUBJ', 'SST', 'HS', 'KS'))
#plt.tight_layout()
ax.legend(bbox_to_anchor=(0.22, 1.0), loc='upper left')
def autolabel(rects, xpos='center'):
"""
Attach a text label above each bar in *rects*, displaying its height.
*xpos* indicates which side to place the text w.r.t. the center of
the bar. It can be one of the following {'center', 'right', 'left'}.
"""
xpos = xpos.lower() # normalize the case of the parameter
ha = {'center': 'center', 'right': 'left', 'left': 'right'}
offset = {'center': 0.5, 'right': 0.53, 'left': 0.47} # x_txt = x + w*off
for rect in rects:
height = rect.get_height()
ax.text(rect.get_x() + rect.get_width() * offset[xpos], 1.01 * height,
'{0:.2f}'.format(height), ha=ha[xpos], va='bottom')
autolabel(rects1, "left")
autolabel(rects2, "center")
autolabel(rects3, "right")
# print os.path.dirname(fold_paths_file) + "/label_switch.pdf"
plt.show()
save_figure_publishable(fig, os.path.dirname(fold_paths_file) + "/label_switch.png")