(args)
| 35 | return all_args |
| 36 | |
| 37 | def test_pytorch(args): |
| 38 | |
| 39 | all_args = parse(args) |
| 40 | |
| 41 | # set all the seed |
| 42 | random.seed(all_args.seed) |
| 43 | torch.manual_seed(all_args.seed) |
| 44 | np.random.seed(all_args.seed) |
| 45 | |
| 46 | run_dir = Path("../results") / all_args.project_name / all_args.experiment_name |
| 47 | if not run_dir.exists(): |
| 48 | os.makedirs(str(run_dir)) |
| 49 | |
| 50 | wandb.init(config=all_args, |
| 51 | project=all_args.project_name, |
| 52 | entity=all_args.team_name, |
| 53 | notes=socket.gethostname(), |
| 54 | name=all_args.experiment_name + "_" + str(all_args.seed), |
| 55 | group=all_args.scenario_name, |
| 56 | dir=str(run_dir), |
| 57 | job_type="training", |
| 58 | reinit=True) |
| 59 | |
| 60 | train_x = np.array([ |
| 61 | [5.0, 3.5, 1.3, 0.3], |
| 62 | [4.5, 2.3, 1.3, 0.3], |
| 63 | [5.5, 2.6, 4.4, 1.2], |
| 64 | [6.1, 3.0, 4.6, 1.4], |
| 65 | [6.7, 3.1, 5.6, 2.4], |
| 66 | [6.9, 3.1, 5.1, 2.3]], dtype=np.float32) |
| 67 | |
| 68 | train_y = np.array([0, 0, 1, 1, 2, 2], dtype=np.long) |
| 69 | |
| 70 | train_x = torch.tensor(train_x, dtype=torch.float32) |
| 71 | train_y = torch.tensor(train_y, dtype=torch.long) |
| 72 | model = Model() |
| 73 | wandb.watch(model,log_freq=1) |
| 74 | max_epochs = 100 |
| 75 | lrn_rate = 0.04 |
| 76 | loss_func = torch.nn.CrossEntropyLoss() |
| 77 | optimizer = torch.optim.SGD(model.parameters(), lr=lrn_rate) |
| 78 | |
| 79 | model.train() |
| 80 | |
| 81 | indices = np.arange(6) |
| 82 | for epoch in range(0, max_epochs): |
| 83 | np.random.shuffle(indices) |
| 84 | for i in indices: |
| 85 | X = train_x[i].reshape(1, 4) |
| 86 | Y = train_y[i].reshape(1, ) |
| 87 | optimizer.zero_grad() |
| 88 | oupt = model(X) |
| 89 | loss_obj = loss_func(oupt, Y) |
| 90 | loss_obj.backward() |
| 91 | optimizer.step() |
| 92 | |
| 93 | print('Epoch:{} Loss:{}'.format(epoch,loss_obj)) |
| 94 |
no test coverage detected