def write_job_script(job_file, job_template, job_params):
'''
Write a job script to job_file by filling in job_template with job_params.
'''
buf = params.format_params(job_params, line_start='# ')
buf = fill_job_template(job_template,
dict(job_name=job_params.name, job_params=buf))
buf = fill_job_template(buf, job_params)
write_file(job_file, buf)
def write_solver(solver_file, solver_params):
'''
Write a dict of solver_params to solver_file.
'''
buf = ''
if solver_params:
buf += params.format_params(solver_params, '# ')
buf += str(make_solver(**solver_params))
write_file(solver_file, buf)
def write_model(model_file, net_param, model_params={}):
'''
Write net_param to model_file. If params dict is given,
write the params as comments in the header of model_file.
'''
buf = ''
if model_params:
buf += params.format_params(model_params, '# ')
buf += str(net_param)
write_file(model_file, buf)
def write_models(model_dir, param_space, scaffold=False, n_benchmark=0, verbose=False):
'''
Write a model in model_dir for every set of params in param_space.
'''
if not os.path.isdir(model_dir):
os.makedirs(model_dir)
if scaffold or n_benchmark > 0:
df = pd.DataFrame(index=range(len(param_space)))
model_names = []
for i, model_params in enumerate(param_space):
model_file = os.path.join(model_dir, '{}.model'.format(model_params.name))
if verbose:
print('CREATING MODEL ' + str(i))
print('model_file = ' + model_file)
print('model_name = ' + model_params.name)
print('model_params = \n' + params.format_params(model_params, ' '), end='')
net_param = make_model(verbose=verbose, **model_params)
write_model(model_file, net_param, model_params)
model_names.append(model_params.name)
if scaffold or n_benchmark > 0:
df.loc[i, 'model_file'] = model_file
net = caffe_util.Net(model_file, caffe.TRAIN)
result = benchmark_net(net, n=n_benchmark)
for key, value in result.mean().items():
df.loc[i, key] = value
if verbose:
print(df.loc[i])
print(model_file)
if scaffold or n_benchmark > 0:
print(df)
print(model_names)