ZhaoKe1024
commented
7 months ago I ran your code and made modifications to the test.py file to print the start and end times of each operations, similar to a Gantt chart. However, I found that the durations of different operations within the same job have overlapping. Specifically, I found that operations within the same job can have overlapping processing times when executed on different machines. In the FJSP problem, the subsequent operation cannot begin until the previous one has finished. I would like to know, if there is an error in my code or if I misunderstood your code. My modified code is as follows, with the addition of a printing section:
def schedule(env, model, memories, flag_sample=False): # Get state and completion signal state = env.state dones = env.done_batch done = False # Unfinished at the beginning biases = env.num_ope_biases_batch[0] proc_time = env.schedules_batch[0] # operation list for every machine res = [[] for _ in range(env.num_mas)] last_time = time.time() i = 0 while ~done: i += 1 with torch.no_grad(): actions = model.policy_old.act(state, memories, dones, flag_sample=flag_sample, flag_train=False) # print("actions:", actions) injob_index = copy.deepcopy(actions[0][0]) # operation global index st = round(proc_time[injob_index][2].item(), 3) # start time et = round(proc_time[injob_index][3].item(), 3) # end time # The following code obtains the index of an operation within the job if injob_index > biases[-1]: injob_index -= biases[-1] for j, k in enumerate(biases): if injob_index > k: continue elif injob_index < k: injob_index -= biases[j-1] break else: if j==0: break else: injob_index = 0 break # print(f"Machine:{actions[1][0]} ; Task({actions[2][0]}-{injob_index})") # Task(job_id - operation_id_in_job)[start_time - end_time] res[actions[1][0]].append(f"Task({actions[2][0]}-{injob_index})[{st},{et}]") # As printed below # Machine0: Task(0-0)[0.0,4.5]||Task(4-1)[5.0,9.0]||Task(10-1)[5.0,7.0]||... # Machine1: Task(1-0)[0.0,6.0]||Task(4-2)[6.0,12.0]||Task(2-0)[0.0,6.0]||Task(3-1)[13.0,19.0]||... # ... state, rewards, dones = env.step(actions) # environment transit done = dones.all() for i, ma in enumerate(res): # print(f"Machine {i}", end=': ') for ta in ma: print(ta, end='||') print() spend_time = time.time() - last_time # The time taken to solve this environment (instance) # print("spend_time: ", spend_time) # Verify the solution gantt_result = env.validate_gantt()[0] if not gantt_result: print("Scheduling Error！！！！！！") return copy.deepcopy(env.makespan_batch), spend_time
In addition, if possible, we can have a detailed discussion. My email address is zhaoke22@mails.ucas.ac.cn , and we can communicate in Chinese.
I ran your code and made modifications to the test.py file to print the start and end times of each operations, similar to a Gantt chart. However, I found that the durations of different operations within the same job have overlapping. Specifically, I found that operations within the same job can have overlapping processing times when executed on different machines. In the FJSP problem, the subsequent operation cannot begin until the previous one has finished. I would like to know, if there is an error in my code or if I misunderstood your code. My modified code is as follows, with the addition of a printing section: