in visualization_utils/plotting.py [0:0]
def plot_codesign_rate_efficacy_cross_workloads_updated_for_paper(input_dir_names, res_column_name_number):
#itrColNum = all_res_column_name_number["iteration cnt"]
#distColNum = all_res_column_name_number["dist_to_goal_non_cost"]
trueNum = all_res_column_name_number["move validity"]
move_name_number = all_res_column_name_number["move name"]
# experiment_names
file_full_addr_list = []
for dir_name in input_dir_names:
file_full_addr = os.path.join(dir_name, "result_summary/FARSI_simple_run_0_1_all_reults.csv")
file_full_addr_list.append(file_full_addr)
axis_font = {'fontname': 'Arial', 'size': '4'}
x_column_name = "iteration cnt"
#y_column_name_list = ["high level optimization name", "exact optimization name", "architectural principle", "comm_comp"]
y_column_name_list = ["exact optimization name", "architectural principle", "comm_comp", "workload"]
#y_column_name_list = ["high level optimization name", "exact optimization name", "architectural principle", "comm_comp"]
column_co_design_cnt = {}
column_non_co_design_cnt = {}
column_co_design_rate = {}
column_non_co_design_rate = {}
column_co_design_efficacy_avg = {}
column_non_co_design_efficacy_rate = {}
column_non_co_design_efficacy = {}
column_co_design_dist= {}
column_co_design_dist_avg= {}
column_co_design_improvement = {}
experiment_name_list = []
last_col_val = ""
ctr_ = 0
for file_full_addr in file_full_addr_list:
if ctr_ == 1:
continue
experiment_name = get_experiments_name(file_full_addr, res_column_name_number)
experiment_name_list.append(experiment_name)
column_co_design_dist_avg[experiment_name] = {}
column_co_design_efficacy_avg[experiment_name] = {}
column_co_design_cnt = {}
for y_column_name in y_column_name_list:
y_column_number = res_column_name_number[y_column_name]
x_column_number = res_column_name_number[x_column_name]
dis_to_goal_column_number = res_column_name_number["dist_to_goal_non_cost"]
ref_des_dis_to_goal_column_number = res_column_name_number["ref_des_dist_to_goal_non_cost"]
column_co_design_cnt[y_column_name] = []
column_non_co_design_cnt[y_column_name] = []
column_non_co_design_efficacy[y_column_name] = []
column_co_design_dist[y_column_name] = []
column_co_design_improvement[y_column_name] = []
column_co_design_rate[y_column_name] = []
all_values = get_all_col_values_of_a_folders(input_dir_names, all_res_column_name_number, y_column_name)
last_row_change = ""
with open(file_full_addr, newline='') as csvfile:
resultReader = csv.reader(csvfile, delimiter=',', quotechar='|')
rows = list(resultReader)
for i, row in enumerate(rows):
if i > 1:
last_row = rows[i - 1]
if row[y_column_number] not in all_values or row[move_name_number]=="identity":
continue
col_value = row[y_column_number]
col_values = col_value.split(";")
for idx, col_val in enumerate(col_values):
# only for improvement
if float(row[ref_des_dis_to_goal_column_number]) - float(row[dis_to_goal_column_number]) < 0:
continue
delta_x_column = (float(row[x_column_number]) - float(last_row[x_column_number]))/len(col_values)
delta_improvement = (float(last_row[dis_to_goal_column_number]) - float(row[dis_to_goal_column_number]))/(float(last_row[dis_to_goal_column_number])*len(col_values))
if not col_val == last_col_val and i > 1:
if not last_row_change == "":
distance_from_last_change = float(last_row[x_column_number]) - float(last_row_change[x_column_number]) + idx * delta_x_column
column_co_design_dist[y_column_name].append(distance_from_last_change)
improvement_from_last_change = (float(last_row[dis_to_goal_column_number]) - float(row[dis_to_goal_column_number]))/float(last_row[dis_to_goal_column_number]) + idx *delta_improvement
column_co_design_improvement[y_column_name].append(improvement_from_last_change)
last_row_change = copy.deepcopy(last_row)
last_col_val = col_val
# co_des cnt
# we ignore the first element as the first element distance is always zero
co_design_dist_sum = 0
co_design_efficacy_sum = 0
avg_ctr = 1
co_design_dist_selected = column_co_design_dist[y_column_name]
co_design_improvement_selected = column_co_design_improvement[y_column_name]
for idx,el in enumerate(column_co_design_dist[y_column_name]):
if idx == len(co_design_dist_selected) - 1:
break
co_design_dist_sum += 1/(column_co_design_dist[y_column_name][idx] + column_co_design_dist[y_column_name][idx+1])
co_design_efficacy_sum += (column_co_design_improvement[y_column_name][idx] + column_co_design_improvement[y_column_name][idx+1])
#/(column_co_design_dist[y_column_name][idx] + column_co_design_dist[y_column_name][idx+1])
avg_ctr+=1
column_co_design_improvement = {}
column_co_design_dist_avg[experiment_name][y_column_name]= co_design_dist_sum/avg_ctr
column_co_design_efficacy_avg[experiment_name][y_column_name] = co_design_efficacy_sum/avg_ctr
ctr_ +=1
#result = {"rate":{}, "efficacy":{}}
#rate_column_co_design = {}
plt.figure()
y_column_name_list_rep = ["L.Opt", "H.Opt", "CM", "WL"]
y_column_name_list_rep_rep = [re.sub("(.{5})", "\\1\n", label, 0, re.DOTALL) for label in y_column_name_list_rep]
plotdata = pd.DataFrame(column_co_design_dist_avg, index=y_column_name_list)
fontSize = 26
# print(plotdata)
# print(y_column_name_list)
color_list={'a0.021_e0.034_h0.034_0.008737_1.7475e-05_arch-aware': 'green', 'a0.021_e0.034_h0.034_0.008737_1.7475e-05_random': 'red'} # Ying: uncomment for blind_study_all_dumb_versions/blind_vs_arch_aware (T.B.M), also the one below
ax = plotdata.plot(kind='bar', fontsize=fontSize, figsize=(6.6, 6.6), color=color_list) # Ying: uncomment for blind_study_all_dumb_versions/blind_vs_arch_aware (T.B.M), also the one below
ax.set_xticklabels(y_column_name_list_rep_rep, rotation=0)
# Ying: hardcode here
ax.set_xlabel("Co-design Parameter", fontsize=fontSize, labelpad=-25)
ax.set_ylabel("Co-design Rate", fontsize=fontSize)
for experiment_name, value in column_co_design_dist_avg.items():
# print(experiment_name[-6:])
if experiment_name[-6:] == "random":
ax.legend(['SA', 'FARSI'], bbox_to_anchor=(0.5, 1.23), loc="upper center", fontsize=fontSize - 2, ncol=2)
else:
ax.legend(['FARSI', 'SA'], bbox_to_anchor=(0.5, 1.23), loc="upper center", fontsize=fontSize - 2, ncol=2)
break
# Ying: hardcode finished
plt.tight_layout()
# dump in the top folder
output_base_dir = '/'.join(input_dir_names[0].split("/")[:-2])
output_dir = os.path.join(output_base_dir, "cross_workloads/co_design_rate")
if not os.path.exists(output_dir):
os.makedirs(output_dir)
plt.savefig(os.path.join(output_dir,"co_design_avg_dist"+'_'.join(y_column_name_list)+".png"), bbox_inches='tight')
# plt.show()
plt.close('all')
plt.figure()
plotdata = pd.DataFrame(column_co_design_efficacy_avg, index=y_column_name_list)
fontSize = 26
ax = plotdata.plot(kind='bar', fontsize=fontSize, figsize=(6.6, 6.6), color=color_list) # Ying: uncomment for blind_study_all_dumb_versions/blind_vs_arch_aware (T.B.M), also the one above
ax.set_xticklabels(y_column_name_list_rep_rep, rotation=0)
# Ying: hardcode here
ax.set_xlabel("Co-design Parameter", fontsize=fontSize, labelpad=-25)
ax.set_ylabel("Co-design Improvement", fontsize=fontSize)
for experiment_name, value in column_co_design_dist_avg.items():
# print(experiment_name[-6:])
if experiment_name[-6:] == "random":
ax.legend(['SA', 'FARSI'], bbox_to_anchor=(0.5, 1.23), loc="upper center", fontsize=fontSize - 2, ncol=2)
else:
ax.legend(['FARSI', 'SA'], bbox_to_anchor=(0.5, 1.23), loc="upper center", fontsize=fontSize - 2, ncol=2)
break
# Ying: hardcode finished
plt.tight_layout()
# dump in the top folder
output_base_dir = '/'.join(input_dir_names[0].split("/")[:-2])
output_dir = os.path.join(output_base_dir, "cross_workloads/co_design_rate")
if not os.path.exists(output_dir):
os.makedirs(output_dir)
plt.savefig(os.path.join(output_dir,"co_design_efficacy"+'_'.join(y_column_name_list)+".png"), bbox_inches='tight')
# plt.show()
plt.close('all')