wangzelin.wzl 93a1074b0c patch 4.0
2022-10-24 17:57:12 +08:00

133 lines
4.1 KiB
Python
Executable File

from mylog.mylog import MyLogger
from op_generator import op_generator
from cost_test_conf import Config
import subprocess as sp
import os
from lmfit import Model
import numpy as np
# step 1 gen op and conf
material_cls = op_generator.gen_operator("material")
conf = Config()
conf.u_to_test_op_c = 'material'
conf.is_not_running_as_unittest_c = True
conf.schema_file_c = 'c10k1.schema'
conf.row_count_c = 1000
conf.input_projector_count_c = 1
material_op = material_cls(conf)
result_file_name = 'material_result'
if os.path.exists(result_file_name):
os.remove(result_file_name)
# step 2 do_bench and gen data
row_count_max = 1001
row_count_step = 100
column_counts = [3, 5, 8]
case_run_time = 7
total_case_count = (row_count_max / row_count_step + 1) * len(column_counts) * case_run_time
case_count = 0
print "Total case count %s ..." % (total_case_count)
for row_count in xrange(1, row_count_max + 1, row_count_step):
for column_count in column_counts:
for time in xrange(case_run_time):
case_count += 1
material_op.conf.row_count_c = row_count
material_op.conf.input_projector_count_c = column_count
sp.check_call("echo -n '%s,' >> %s" % (row_count, result_file_name), shell=True)
sp.check_call("echo -n '%s,' >> %s" % (column_count, result_file_name), shell=True)
print "Running case %s / %s ... : %s " % (case_count, total_case_count, material_op.get_bench_cmd())
print "%s >> %s" % (material_op.get_bench_cmd(), result_file_name)
sp.check_call("%s >> %s" % (material_op.get_bench_cmd(), result_file_name), shell=True)
# step 3 preprocess data
final_file_name = "material_result_final"
if os.path.exists("material_final_result"):
os.remove("material_final_result")
data_cmd = material_op.get_data_preprocess_cmd()
sp.check_call(data_cmd, shell=True)
# step 4 fit and output
# given model form, do fit using previous result data
# case param should be considered with cost_model_util.cpp output format
# eg: material_test() in cost_model_util.cpp
# output row_count, cost_time
out_model_file_name = "material_model"
if os.path.exists(out_model_file_name):
os.remove(out_model_file_name)
def material_model_form(args,
# Tstartup,
Trow_once,
Trow_col):
(
Nrow,
Ncol,
) = args
total_cost = 0 # Tstartup
total_cost += Nrow * (Trow_once + Ncol * Trow_col)
return total_cost
def material_model_arr(arg_sets,
# Tstartup,
Trow_once,
Trow_col):
res = []
for single_arg_set in arg_sets:
res.append(material_model_form(single_arg_set,
# Tstartup,
Trow_once,
Trow_col))
return np.array(res)
def extract_info_from_line(line):
splited = line.split(",")
line_info = []
for item in splited:
line_info.append(float(item))
return line_info
material_model = Model(material_model_arr)
material_model.set_param_hint("Trow_once", min=0.0)
material_model.set_param_hint("Trow_col", min=0.0)
file = open(final_file_name, "r")
arg_sets = []
times = []
case_params = []
for line in file:
if line.startswith('#'):
continue
case_param = extract_info_from_line(line)
case_params.append(case_param)
arg_sets.append((case_param[0], case_param[1]))
times.append(case_param[3])
file.close()
arg_sets_np = np.array(arg_sets)
times_np = np.array(times)
# result is the fitting result model
result = material_model.fit(times_np, arg_sets=arg_sets_np,
# Tstartup=10.0,
Trow_once=10.0,
Trow_col=1.0
)
# res_line = str(result.best_values["Tstartup"]) + ","
res_line = str(result.best_values["Trow_once"]) + ","
res_line += str(result.best_values["Trow_col"])
print result.fit_report()
if out_model_file_name:
out_file = open(out_model_file_name, "w")
out_file.write(res_line)
out_file.close()