提交 4b014a1b 编写于 作者: W weishengyu

dbg

上级 45f2d010
...@@ -23,13 +23,16 @@ search_dict: ...@@ -23,13 +23,16 @@ search_dict:
search_values: [0.0, 0.1, 0.5] search_values: [0.0, 0.1, 0.5]
lr_mult_list: lr_mult_list:
replace_config: replace_config:
- Arch.models.1.Student.lr_mult_list - Arch.lr_mult_list
search_values: search_values:
- [0.0, 0.2, 0.4, 0.6, 0.8, 1.0] - [0.0, 0.2, 0.4, 0.6, 0.8, 1.0]
- [0.0, 0.4, 0.4, 0.8, 0.8, 1.0] - [0.0, 0.4, 0.4, 0.8, 0.8, 1.0]
teacher: teacher:
rm_keys: rm_keys:
- Arch.models.1.Student.lr_mult_list - Arch.lr_mult_list
search_values: search_values:
- ResNet101_vd - ResNet101_vd
- ResNet50_vd - ResNet50_vd
final_replace:
Arch.lr_mult_list: Arch.models.1.Student.lr_mult_list
...@@ -33,7 +33,6 @@ Arch: ...@@ -33,7 +33,6 @@ Arch:
- Teacher: - Teacher:
name: ResNet101_vd name: ResNet101_vd
class_num: *class_num class_num: *class_num
pretrained: "./output/TEACHER_ResNet101_vd/ResNet101_vd/best_model"
- Student: - Student:
name: PPLCNet_x1_0 name: PPLCNet_x1_0
class_num: *class_num class_num: *class_num
......
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(os.path.abspath(os.path.join(__dir__, '../')))
import subprocess import subprocess
from ppcls.utils import config from ppcls.utils import config
...@@ -6,11 +14,11 @@ def get_result(log_dir): ...@@ -6,11 +14,11 @@ def get_result(log_dir):
log_file = "{}/train.log".format(log_dir) log_file = "{}/train.log".format(log_dir)
with open(log_file, "r") as f: with open(log_file, "r") as f:
raw = f.read() raw = f.read()
res = float(raw.split("best metric ")[-1].split("]")[0]) res = float(raw.split("best metric: ")[-1].split("]")[0])
return res return res
def search_train(search_list, base_program, base_output_dir, search_key, config_replace_value): def search_train(search_list, base_program, base_output_dir, search_key, config_replace_value, model_name):
best_res = 0. best_res = 0.
best = search_list[0] best = search_list[0]
all_result = {} all_result = {}
...@@ -18,11 +26,14 @@ def search_train(search_list, base_program, base_output_dir, search_key, config_ ...@@ -18,11 +26,14 @@ def search_train(search_list, base_program, base_output_dir, search_key, config_
program = base_program.copy() program = base_program.copy()
for v in config_replace_value: for v in config_replace_value:
program += ["-o", "{}={}".format(v, search_i)] program += ["-o", "{}={}".format(v, search_i)]
output_dir = "{}/{}_{}".format(base_output_dir, search_key, search_i.replace(".", "_")) if v == "Arch.name":
model_name = search_i
output_dir = "{}/{}_{}".format(base_output_dir, search_key, search_i).replace(".", "_")
program += ["-o", "Global.output_dir={}".format(output_dir)] program += ["-o", "Global.output_dir={}".format(output_dir)]
subprocess.Popen(program) process = subprocess.Popen(program)
res = get_result(output_dir) process.communicate()
all_result[search_i] = res res = get_result("{}/{}".format(output_dir, model_name))
all_result[str(search_i)] = res
if res > best_res: if res > best_res:
best = search_i best = search_i
best_res = res best_res = res
...@@ -34,16 +45,19 @@ def search_strategy(): ...@@ -34,16 +45,19 @@ def search_strategy():
args = config.parse_args() args = config.parse_args()
configs = config.get_config(args.config, overrides=args.override, show=False) configs = config.get_config(args.config, overrides=args.override, show=False)
base_config_file = configs["base_config_file"] base_config_file = configs["base_config_file"]
distill_config_file = configs["distill_config_file"]
model_name = config.get_config(base_config_file)["Arch"]["name"]
gpus = configs["gpus"] gpus = configs["gpus"]
gpus = ",".join([str(i) for i in gpus])
base_program = ["python3.7", "-m", "paddle.distributed.launch", "--gpus={}".format(gpus), base_program = ["python3.7", "-m", "paddle.distributed.launch", "--gpus={}".format(gpus),
"tools/train.py", "-c", base_config_file] "tools/train.py", "-c", base_config_file]
base_output_dir = configs["output_dir"] base_output_dir = configs["output_dir"]
search_dict = configs.get("search_dict") search_dict = configs.get("search_dict")
all_results = {} all_results = {}
for search_key in search_dict: for search_key in search_dict:
search_values = configs[search_key]["search_values"] search_values = search_dict[search_key]["search_values"]
replace_config = search_dict[search_key]["replace_config"] replace_config = search_dict[search_key]["replace_config"]
res = search_train(search_values, base_program, base_output_dir, search_key, replace_config) res = search_train(search_values, base_program, base_output_dir, search_key, replace_config, model_name)
all_results[search_key] = res all_results[search_key] = res
best = res.get("best") best = res.get("best")
for v in replace_config: for v in replace_config:
...@@ -56,22 +70,33 @@ def search_strategy(): ...@@ -56,22 +70,33 @@ def search_strategy():
teacher_rm_keys = teacher_configs["rm_keys"] teacher_rm_keys = teacher_configs["rm_keys"]
rm_indices = [] rm_indices = []
for rm_k in teacher_rm_keys: for rm_k in teacher_rm_keys:
rm_indices.append(base_program.index(rm_k)) for ind, ki in enumerate(base_program):
rm_indices = sorted(rm_indices) if rm_k in ki:
for rm_index in rm_indices[:, :, -1]: rm_indices.append(ind)
teacher_program.pop(rm_index + 1) print(rm_indices)
for rm_index in rm_indices[::-1]:
teacher_program.pop(rm_index) teacher_program.pop(rm_index)
replace_config = "-o Arch.name" teacher_program.pop(rm_index-1)
replace_config = ["Arch.name"]
teacher_list = teacher_configs["search_values"] teacher_list = teacher_configs["search_values"]
res = search_train(teacher_list, teacher_program, base_output_dir, "teacher", replace_config) res = search_train(teacher_list, teacher_program, base_output_dir, "teacher", replace_config, model_name)
all_results["teacher"] = res all_results["teacher"] = res
best = res.get("best") best = res.get("best")
t_pretrained = "{}/{}_{}".format(base_output_dir, "teacher", best.replace(".", "_")) t_pretrained = "{}/{}_{}/{}/best_model".format(base_output_dir, "teacher", best, best)
base_program += ["-o", "Arch.models.0.Teacher.name={}".format(best), base_program += ["-o", "Arch.models.0.Teacher.name={}".format(best),
"-o", "Arch.models.0.Teacher.pretrained={}".format(t_pretrained)] "-o", "Arch.models.0.Teacher.pretrained={}".format(t_pretrained)]
output_dir = "{}/search_res".format(base_output_dir) output_dir = "{}/search_res".format(base_output_dir)
base_program += ["-o", "Global.output_dir={}".format(output_dir)] base_program += ["-o", "Global.output_dir={}".format(output_dir)]
final_replace = configs.get('final_replace')
for i in range(len(base_program)):
base_program[i] = base_program[i].replace(base_config_file, distill_config_file)
for k in final_replace:
v = final_replace[k]
base_program[i] = base_program[i].replace(k, v)
subprocess.Popen(base_program) subprocess.Popen(base_program)
subprocess.communicate()
print(all_results, base_program)
if __name__ == '__main__': if __name__ == '__main__':
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册