atomest/trainer.py at main · ScaleMind-C9308A/atomest · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
from ds import get_data_numpy, get_data_tgeo
from models import get_model

from sklearn.metrics import mean_squared_error, median_absolute_error

from torch_geometric.loader import DataLoader

from torch.optim import Adam
from torch.nn import MSELoss, L1Loss

import numpy as np
import json
import hashlib
import os
import torch


def get_hash(args):
    args_str = json.dumps(vars(args), sort_keys=True)
    args_hash = hashlib.md5(args_str.encode('utf-8')).hexdigest()
    return args_hash

def save_json(dct, path):
    with open(path, 'w') as outfile:
        json.dump(dct, outfile)

def read_json(path):
    return json.load(open(path, 'r'))

def folder_setup(args):

    run_name = get_hash(args)

    run_dir = os.getcwd() + "/runs"
    if not os.path.exists(run_dir):
        os.mkdir(run_dir)

    save_dir = run_dir + f"/{run_name}"
    if not os.path.exists(save_dir):
        os.mkdir(save_dir)

    save_json(vars(args), save_dir + f"/config.json")

    return save_dir


def ml_train(args):

    save_dir = folder_setup(args)

    Xs, Ts, scale = get_data_numpy()

    print(Xs.shape, Ts.shape)
    print(Xs.min(), Ts.min())
    print(Xs.max(), Ts.max())

    fold_cnt = Xs.shape[0]

    log = {}

    for fold_idx in range(fold_cnt):

        train_idx = list(range(fold_cnt))
        train_idx.remove(fold_idx)
        valid_idx = fold_idx


        X_train = np.concatenate(Xs[train_idx, :])
        T_train = np.concatenate(Ts[train_idx, :])

        X_valid = Xs[valid_idx, :]
        T_valid = Ts[valid_idx, :]

        model = get_model(args)

        model.fit(X_train, T_train)
        train_score = model.score(X_train, T_train)
        valid_score = model.score(X_valid, T_valid)

        print(f"Score - {fold_idx}", train_score, valid_score)

        T_tpred = model.predict(X_train)
        T_vpred = model.predict(X_valid)

        train_mse = mean_squared_error(T_tpred, T_train).item()
        valid_mse = mean_squared_error(T_vpred, T_valid).item()

        print(f"MSE - {fold_idx}", train_mse, valid_mse)

        train_mae = median_absolute_error(T_tpred, T_train).item()
        valid_mae = median_absolute_error(T_vpred, T_valid).item()

        print(f"MAE - {fold_idx}", train_mae, valid_mae)

        log[fold_idx] = {'train_mse' : train_mse, 'train_mae' : train_mae, 'valid_mse' : valid_mse, 'valid_mae' : valid_mae}

    save_json(log, save_dir + "/results.json")


def graph_train(args):

    save_dir = folder_setup(args)

    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

    criterion = MSELoss()
    metric = L1Loss()

    folds, scale = get_data_tgeo()

    print(f"#Folds: {len(folds)}")

    fold_cnt = len(folds)

    log = {x : [] for x in range(fold_cnt)}

    for fold_idx in range(fold_cnt):

        valid_graphs = folds[fold_idx]
        train_graphs = sum([x for i, x in enumerate(folds) if i != fold_idx], [])

        train_ld = DataLoader(train_graphs, batch_size=args.bs, shuffle=True)
        valid_ld = DataLoader(valid_graphs, batch_size=args.bs, shuffle=True)

        print(f"Number Train Batch: {len(train_ld)}")
        print(f"Number Valid Batch: {len(valid_ld)}")

        model = get_model(args).to(device)
        optimizer = Adam(model.parameters(), lr=args.lr)

        for epoch in range(args.epoch):
            print(f"Epoch: {epoch}")
            train_total_mse = 0
            train_total_mae = 0
            model.train()
            for data in train_ld:
                data = data.to(device)
                out = model(data.x, data.edge_index, data.edge_attr, data.batch)
                loss = criterion(out, data.y)
                optimizer.zero_grad()
                loss.backward()
                optimizer.step()

                train_total_mse += criterion(out, data.y).item() / len(train_ld)
                train_total_mae += metric(out, data.y).item() / len(train_ld)

            model.eval()
            with torch.no_grad():
                valid_total_mse = 0
                valid_total_mae = 0
                for data in valid_ld:
                    data = data.to(device)
                    out = model(data.x, data.edge_index, data.edge_attr, data.batch)
                    loss = criterion(out, data.y)

                    valid_total_mse += criterion(out, data.y).item() / len(valid_ld)
                    valid_total_mae += metric(out, data.y).item() / len(valid_ld)

            print(f"\tMSE - {fold_idx}", train_total_mse, valid_total_mse)
            print(f"\tMAE - {fold_idx}", train_total_mae, valid_total_mae)

            log[fold_idx].append({'train_mse' : train_total_mse, 'train_mae' : train_total_mae, 'valid_mse' : valid_total_mse, 'valid_mae' : valid_total_mae})


    save_json(log, save_dir + "/results.json")