def main():
#Load the model from architecture
model, optimizer = Initialize_Att_model(args)
model.eval()
args.gpu = False
###make SWA name
model_name = str(args.model_dir).split('/')[-1]
ct = model_name + '_SWA_random_tag_' + str(args.SWA_random_tag)
##check the Weight averaged file and if the file does not exist then lcreate them
## if the file exists load them
if not isfile(join(args.model_dir, ct)):
model_names, checkpoint_ter = get_best_weights(args.weight_text_file,
args.Res_text_file)
model_names_checkpoints = model_names[:args.early_stopping_checkpoints]
model = Stocasting_Weight_Addition(model, model_names_checkpoints)
torch.save(model.state_dict(), join(args.model_dir, ct))
else:
print("taking the weights from", ct, join(args.model_dir, str(ct)))
args.pre_trained_weight = join(args.model_dir, str(ct))
model, optimizer = Initialize_Att_model(args)
#---------------------------------------------
model.eval()
print("best_weight_file_after stocastic weight averaging")
#=================================================
model = model.cuda() if args.gpu else model
plot_path = join(args.model_dir, 'decoding_files', 'plots')
#=================================================
#=================================================
####read all the scps and make large scp with each lines as a feature
decoding_files_list = glob.glob(args.dev_path + "*")
scp_paths_decoding = []
for i_scp in decoding_files_list:
scp_paths_decoding_temp = open(i_scp, 'r').readlines()
scp_paths_decoding += scp_paths_decoding_temp
#scp_paths_decoding this should contain all the scp files for decoding
#====================================================
###sometime i tend to specify more jobs than maximum number of lines in that case python indexing error we get
job_no = int(args.Decoding_job_no) - 1
#args.gamma=0.5
#print(job_no)
#####get_cer_for_beam takes a list as input
present_path = [scp_paths_decoding[job_no]]
text_file_dict = {
line.split(' ')[0]: line.strip().split(' ')[1:]
for line in open(args.text_file)
}
get_cer_for_beam(present_path, model, text_file_dict, plot_path, args)
def main():
args.gpu = False
best_weight_file = get_best_weights(args.weight_text_file,
args.Res_text_file)
print("best_weight_file", best_weight_file)
args.pre_trained_weight = join(best_weight_file)
#=================================================
model, optimizer = Initialize_Att_model(args)
model.eval()
model = model.cuda() if args.gpu else model
#=================================================
plot_path = join(args.model_dir, 'decoding_files', 'plots')
if not isdir(plot_path):
os.makedirs(plot_path)
#=================================================
####read all the scps and make large scp with each lines as a feature
decoding_files_list = glob.glob(args.dev_path + "*")
scp_paths_decoding = []
for i_scp in decoding_files_list:
scp_paths_decoding_temp = open(i_scp, 'r').readlines()
scp_paths_decoding += scp_paths_decoding_temp
#scp_paths_decoding this should contain all the scp files for decoding
#====================================================
###sometime i tend to specify more jobs than maximum number of lines in that case python indexing error we get
job_no = int(args.Decoding_job_no) - 1
#args.gamma=0.5
#print(job_no)
#####get_cer_for_beam takes a list as input
present_path = [scp_paths_decoding[job_no]]
#print(present_path)
text_file_dict = {
line.split(' ')[0]: line.strip().split(' ')[1:]
for line in open(args.text_file)
}
get_cer_for_beam(present_path, model, text_file_dict, plot_path, args)