rnaglib-experiments/layers.py at main · BorgwardtLab/rnaglib-experiments · 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
"""experiment setup."""

import os
import sys

from rnaglib.learning.task_models import PygModel
from rnaglib.tasks import get_task
from rnaglib.transforms import GraphRepresentation
from rnaglib.config.graph_keys import GRAPH_KEYS, TOOL

current_dir = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, os.path.join(current_dir, ".."))
from exp import RNATrainer

MODEL_ARGS = {
    "rna_cm": {
        "2.5D":{
            "hidden_channels": 128
        },
        "2D":{
            "hidden_channels": 128
        },
        "2D_GCN":{
            "hidden_channels": 128
        }
    },
    "rna_prot": {
        "2.5D":{
            "hidden_channels": 64,
            "dropout_rate": 0.2
        },
        "2D":{
            "hidden_channels": 64,
            "dropout_rate": 0.2
        },
        "2D_GCN":{
            "hidden_channels": 64,
            "dropout_rate": 0.2
        },
    },
    "rna_site": {
        "2.5D":{
            "hidden_channels": 256
        },
        "2D":{
            "hidden_channels":128
        },
        "2D_GCN":{
            "hidden_channels": 128
        }
    },
}

# There are only marginal improvements running a hundred epochs, so we leave it at 40 for the splitting analysis
TRAINER_ARGS = {
    "rna_cm": {
        "2.5D":{
            "epochs": 40,
            "batch_size": 8,
            "learning_rate": 0.001
        },
        "2D":{
            "epochs": 40,
            "batch_size": 8,
            "learning_rate": 0.0001
        }
    },
    "rna_prot": {
        "2.5D":{
            "epochs": 40,
            "batch_size": 8,
            "learning_rate": 0.01
        },
         "2D":{
            "epochs": 40,
            "batch_size": 8,
            "learning_rate": 0.01
        },
         "2D_GCN":{
            "epochs": 40,
            "batch_size": 8,
            "learning_rate": 0.01
        },
    },  # 0.01 (original)
    "rna_site": {
        "2.5D":{
            "batch_size": 8,
            "epochs": 40,
            "learning_rate": 0.001
        },
        "2D":{
            "batch_size": 8,
            "epochs": 40,
            "learning_rate": 0.0001
        },
        "2D_GCN":{
            "batch_size": 8,
            "epochs": 40,
            "learning_rate": 0.0001
        }
    },
}

edge_maps = {
    "2.5D": "edge_map",
    "simplified_2.5D": "simplified_edge_map",
    "2D": "2D_edge_map"
}
representation = "2D"
layer_type = "gcn"

if __name__ == "__main__":
    for ta_name in ["rna_site", "rna_cm", "rna_prot"]:

        ta = get_task(root="roots/" + ta_name, task_id=ta_name)
        edge_map = GRAPH_KEYS[edge_maps[representation]][TOOL]
        rep = GraphRepresentation(framework="pyg", edge_map=edge_map)
        ta.dataset.add_representation(rep)
        for seed in [0, 1, 2]:
            ta.get_split_loaders(batch_size=TRAINER_ARGS[ta_name][representation]["batch_size"], recompute=True)
            for nb_layers in [2, 3, 4, 5, 6]:
                if representation=="2D" and layer_type=="gcn":
                    complete_representation_name = "2D_GCN"
                else:
                    complete_representation_name = representation
                exp_name = (f"{ta_name}_{complete_representation_name}_{nb_layers}layers_lr{TRAINER_ARGS[ta_name][representation]['learning_rate']}_"
                            f"{TRAINER_ARGS[ta_name][representation]['epochs']}epochs_hiddendim{MODEL_ARGS[ta_name][representation]['hidden_channels']}_"
                            f"batch_size{TRAINER_ARGS[ta_name][representation]['batch_size']}")
                model = PygModel(num_node_features=ta.metadata["num_node_features"],
                                 num_classes=ta.metadata["num_classes"],
                                 graph_level=ta.metadata["graph_level"],
                                 num_layers=nb_layers, **MODEL_ARGS[ta_name][representation])
                trainer = RNATrainer(ta, model, rep, exp_name=exp_name + "_seed" + str(seed),
                                     learning_rate=TRAINER_ARGS[ta_name][representation]["learning_rate"],
                                     epochs=TRAINER_ARGS[ta_name][representation]["epochs"], seed=seed,
                                     batch_size=TRAINER_ARGS[ta_name][representation]["batch_size"])
                trainer.train()