Selective_SSM_for_EHR_Classification/cli.py at main · gsn245/Selective_SSM_for_EHR_Classification · 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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
from mortality_part_preprocessing import load_pad_separate
from mortality_classification import train_test
import os
import click
import torch
import random
import numpy as np
import json


@click.command()
@click.option(
    "--output_path",
    default="./ehr_classification_results/",
    help="Path to output folder",
)
@click.option("--pooling", default="max", help="pooling function")
@click.option("--epochs", default=300, help="model dropout rate")
@click.option("--dropout", default=0.4, help="model dropout rate")
@click.option("--attn_dropout", default=0.4, help="model attention dropout rate")
@click.option(
    "--model_type", default="transformer", help="model_type"
)
@click.option("--heads", default=1, help="number of attention heads")
@click.option("--batch_size", default=64, help="batch size")
@click.option("--layers", default=1, help="number of attention layers")
@click.option("--dataset_id", default="physionet2012", help="filename id of dataset")
@click.option("--base_path", default="./P12data", help="Path to data folder")
@click.option("--lr", default=0.001, help="learning rate")
@click.option("--patience", default=10, help="patience for early stopping")
@click.option(
    "--use_mask",
    default=False,
    help="boolean, use mask for timepoints with no measurements",
)
@click.option(
    "--early_stop_criteria",
    default="auroc",
    help="what to early stop on. Options are: auroc, auprc, auprc+auroc, or loss",
)
@click.option("--seft_n_phi_layers", default=3)
@click.option("--seft_phi_width", default=32)
@click.option("--seft_phi_dropout", default=0.)
@click.option("--seft_n_psi_layers", default=2)
@click.option("--seft_psi_width", default=64)
@click.option("--seft_psi_latent_width", default=128)
@click.option("--seft_dot_prod_dim", default=128)
@click.option("--seft_latent_width", default=128)
@click.option("--seft_n_rho_layers", default=3)
@click.option("--seft_rho_width", default=32)
@click.option("--seft_rho_dropout", default=0.)
@click.option("--seft_max_timescales", default=100)
@click.option("--seft_n_positional_dims", default=4)
@click.option("--ipnets_imputation_stepsize", default=0.25)
@click.option("--ipnets_reconst_fraction", default=0.25)
@click.option("--recurrent_dropout", default=0.3)
@click.option("--recurrent_n_units", default=100)
def core_function(
    output_path,
    base_path,
    model_type,
    epochs,
    dataset_id,
    batch_size,
    lr,
    patience,
    early_stop_criteria,
    **kwargs
):

    model_args = kwargs

    torch.manual_seed(0)
    random.seed(0)
    np.random.seed(0)

    accum_loss = []
    accum_accuracy = []
    accum_auprc = []
    accum_auroc = []

    for split_index in range(1, 6):

        base_path_new = f"{base_path}/split_{split_index}"
        train_pair, val_data, test_data = load_pad_separate(
            dataset_id, base_path_new, split_index
        )

        # make necessary folders
        # if new model, make model folder
        if os.path.exists(output_path):
            pass
        else:
            try:
                os.mkdir(output_path)
            except OSError as err:
                print("OS error:", err)
        # make run folder
        base_run_path = os.path.join(output_path, f"split_{split_index}")
        run_path = base_run_path
        if os.path.exists(run_path):
            raise ValueError(f"Path {run_path} already exists.")
        os.mkdir(run_path)

        # save model settings
        model_settings = {
            "model_type": model_type,
            "batch_size": batch_size,
            "epochs": epochs,
            "dataset": dataset_id,
            "learning_rate": lr,
            "patience": patience,
            "early_stop_criteria": early_stop_criteria,
            "base_path": base_path,
            "pooling_fxn": model_args["pooling"],
        }
        if model_type == "transformer":
            model_settings["layers"] = model_args["layers"]
        if model_type in ("seft", "transformer"):
            model_settings["dropout"] = model_args["dropout"]
            model_settings["attn_dropout"] = model_args["attn_dropout"]
            model_settings["use_timepoint_mask"] = model_args["use_mask"]
            model_settings["heads"] = model_args["heads"]
        if model_type == "seft":
            model_settings["seft_n_phi_layers"] = model_args["seft_n_phi_layers"]
            model_settings["seft_phi_width"] = model_args["seft_phi_width"]
            model_settings["seft_phi_dropout"] = model_args["seft_phi_dropout"]
            model_settings["seft_n_psi_layers"] = model_args["seft_n_psi_layers"]
            model_settings["seft_psi_width"] = model_args["seft_psi_width"]
            model_settings["seft_psi_latent_width"] = model_args["seft_psi_latent_width"]
            model_settings["seft_dot_prod_dim"] = model_args["seft_dot_prod_dim"]
            model_settings["seft_latent_width"] = model_args["seft_latent_width"]
            model_settings["seft_n_rho_layers"] = model_args["seft_n_rho_layers"]
            model_settings["seft_rho_width"] = model_args["seft_rho_width"]
            model_settings["seft_rho_dropout"] = model_args["seft_rho_dropout"]
        if model_type in ("grud", "ipnets"):
            model_settings["recurrent_dropout"] = model_args["recurrent_dropout"]
            model_settings["recurrent_n_units"] = model_args["recurrent_n_units"]
        if model_type == "ipnets":
            model_settings["ipnets_imputation_stepsize"] = model_args["ipnets_imputation_stepsize"]
            model_settings["ipnets_reconst_fraction"] = model_args["ipnets_reconst_fraction"]

        with open(f"{run_path}/model_settings.json", "w") as fp:
            json.dump(model_settings, fp)

        # run training
        loss, accuracy_score, auprc_score, auc_score = train_test(
            train_pair,
            val_data,
            test_data,
            output_path=run_path,
            model_type=model_type,
            epochs=epochs,
            batch_size=batch_size,
            lr=lr,
            patience=patience,
            early_stop_criteria=early_stop_criteria,
            model_args=model_args,
        )

        accum_loss.append(loss)
        accum_accuracy.append(accuracy_score)
        accum_auprc.append(auprc_score)
        accum_auroc.append(auc_score)

    with open(f"{output_path}/summary.json", "w") as f:
        json.dump(
            {
                "mean_loss": float(np.mean(accum_loss)),
                "mean_accuracy": float(np.mean(accum_loss)),
                "mean_auprc": float(np.mean(accum_loss)),
                "mean_auroc": float(np.mean(accum_loss)),
                "std_loss": float(np.std(accum_loss)),
                "std_accuracy": float(np.std(accum_loss)),
                "std_auprc": float(np.std(accum_loss)),
                "std_auroc": float(np.std(accum_loss)),
            }, f, indent=4,
        )


if __name__ == "__main__":
    core_function()