l = round(random.uniform(1e-8, lrini),8) if l not in lrs: lrs.append(l) print(lrs, len(lrs)) mse_train = [] mse_val = [] for i in range(300): hp = {'epochs': 2000, 'batchsize': int(bs), 'lr': lrs[i], 'eta': eta} print("Hyperparameters", hp) data_dir = "./data/" data = "reg" loss = training.finetune(hp, model_design, (train_x, train_y), (test_x, test_y), data_dir, data, emb=False, reg=(train_yp, test_yp)) mse_train.append(np.mean(loss['train_loss'])) mse_val.append(np.mean(loss['val_loss'])) df = pd.DataFrame(lrs) df['train_loss'] = mse_train df['val_loss'] = mse_val print("Random hparams search best result:") print(df.loc[[df["val_loss"].idxmin()]]) lr = lrs[df["val_loss"].idxmin()] print("Dataframe:", df) df.to_csv("reg_lr.csv")
#print(len(x), len(y)) #print(splits) train_x.index, train_y.index = np.arange(0, len(train_x)), np.arange( 0, len(train_y)) test_x.index, test_y.index = np.arange(0, len(test_x)), np.arange(0, len(test_y)) print("train_x", train_x, test_x) model_design = {'layersizes': [256]} hp = {'epochs': 100000, 'batchsize': int(128), 'lr': 0.01} print(hp) print("TRAIN_TEST", train_x.shape, test_x.shape, "END") data_dir = "./data/" data = "embof" tloss = training.finetune(hp, model_design, (train_x, train_y), (test_x, test_y), data_dir, data) #pd.DataFrame.from_dict(tloss).to_csv('res2_test.csv') print(tloss) train_loss = tloss['train_loss'] val_loss = tloss['val_loss'] pd.DataFrame({ "train_loss": train_loss, "val_loss": val_loss }).to_csv('OFres_vloss.csv')
test_x.index, test_y.index, yp_test.index = np.arange( 0, len(test_x)), np.arange(0, len(test_y)), np.arange(0, len(yp_test)) print("train_x", test_y, yp_test) print("SIZES", train_x, train_y, yp_train) model_design = {'layersizes': [128]} hp = {'epochs': 10000, 'batchsize': int(128), 'lr': 0.01} print(hp) print("TRAIN_TEST", train_x.shape, test_x.shape, "END") data_dir = "./data/" data = "res2of" tloss = training.finetune(hp, model_design, (train_x, train_y), (test_x, test_y), data_dir, data, res=2, ypreles=(yp_train, yp_test)) #pd.DataFrame.from_dict(tloss).to_csv('res2_test.csv') print(tloss) train_loss = tloss['train_loss'] val_loss = tloss['val_loss'] pd.DataFrame({ "train_loss": train_loss, "val_loss": val_loss }).to_csv('OFres2_vloss.csv')
print(lrs, len(lrs)) mse_train = [] mse_val = [] print("trainshape", train_x.shape, train_y.to_frame().shape) for i in range(300): hp = {'epochs': 2000, 'batchsize': int(bs), 'lr': lrs[i]} data_dir = "./data/" data = "2res" loss = training.finetune(hp, model_design, (train_x, train_y.to_frame()), (test_x, test_y.to_frame()), data_dir, data, reg=None, emb=False) mse_train.append(np.mean(loss['train_loss'])) mse_val.append(np.mean(loss['val_loss'])) df = pd.DataFrame(lrs) df['train_loss'] = mse_train df['val_loss'] = mse_val print("Random hparams search best result:") print(df.loc[[df["val_loss"].idxmin()]]) lr = lrs[df["val_loss"].idxmin()] print("Dataframe:", df) df.to_csv("2res_lr.csv")
print("train_x", train_x, rtr) model_design = {'layersizes': [[128], [128]]} hp = {'epochs': 1000, 'batchsize': int(128), 'lr': 0.01, 'eta': 0.5} print(hp) print("TRAIN_TEST", train_x.shape, test_x.shape, "END") data_dir = "./data/" data = "embof" tloss = training.finetune(hp, model_design, (train_x, train_y), (test_x, test_y), data_dir, data, reg=(yp_tr, yp_te), raw=(rtr, rte), emb=True, sw=(swmn, swstd), embtp=2, qn=True) #pd.DataFrame.from_dict(tloss).to_csv('res2_test.csv') print(tloss) train_loss = tloss['train_loss'] val_loss = tloss['val_loss'] pd.DataFrame({ "train_loss": train_loss, "val_loss": val_loss }).to_csv('OFemb_vloss.csv')
print(lrs, len(lrs)) mse_train = [] mse_val = [] for i in range(100): hp = {'epochs': 500, 'batchsize': int(bs), 'lr': lrs[i], 'eta': eta} data_dir = "./data/" data = "emb2" loss = training.finetune(hp, model_design, (train_x, train_y), (test_x, test_y), data_dir, data, reg=(train_yp, test_yp), raw=(train_xr, test_xr), emb=True, sw=(swmn, swstd), embtp=2, exp=2) mse_train.append(np.mean(loss['train_loss'])) mse_val.append(np.mean(loss['val_loss'])) df = pd.DataFrame(lrs) df['train_loss'] = mse_train df['val_loss'] = mse_val print("Random hparams search best result:") print(df.loc[[df["val_loss"].idxmin()]]) lr = lrs[df["val_loss"].idxmin()] print("Dataframe:", df)