Spaces:
Running on Zero
Running on Zero
File size: 10,536 Bytes
093748c | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 | import torch
import os
from torch_geometric.loader import DataLoader
from pathlib import Path
from machine_learning.EarlyStopping import EarlyStopping
from machine_learning.Dataset import BindingResidueDatasetWithLabels
from machine_learning.ModelManager import initialize_untrained_model, save_classifier_torch
from machine_learning.Evaluator import BindingResiduePredictionEvaluator
from setup.configProcessor import get_cv_splits, get_optimizer_arguments, get_weight_dir, do_logging, \
get_model_parameter_dict
from setup.generalSetup import select_device
def run_optimization(model_parameters):
model_counter = 0
performances = []
for structure_cutoff in model_parameters["cutoff_structure"]:
for nr_feature_channels in model_parameters["features"]:
for dropout in model_parameters["dropout"]:
for batch_size in model_parameters["batchsize"]:
for epochs in model_parameters["epochs"]:
for weights in model_parameters["weights"]:
for dropout_fn in model_parameters["droupout_fcn"]:
model_parameters_run = {
"cutoff": model_parameters["cutoff"],
"in_channels": model_parameters["in_channels"],
"out_channels": model_parameters["out_channels"],
"features": nr_feature_channels,
"dropout": dropout,
"epochs": epochs,
"early_stopping": model_parameters["early_stopping"],
"weights": weights,
"cutoff_structure": structure_cutoff,
"activation": model_parameters["activation"],
"batchsize": batch_size,
"droupout_fcn": dropout_fn,
"use_dssp": model_parameters["use_dssp"],
"nr_dssp_features": model_parameters["nr_dssp_features"]
}
print(f"-----------Training model{model_counter}-----------")
training_evaluators, validation_evaluators = \
run_training(model_parameters_run, filename_model_prefix=f"model{model_counter}_")
training_performances = [training_evaluator.performances["all"]["f1"][-1]
for training_evaluator in training_evaluators]
validation_performances = [validation_evaluator.performances["all"]["f1"][-1]
for validation_evaluator in validation_evaluators]
average_training_performance = sum(training_performances)/len(training_performances)
average_validation_performance = \
sum(validation_performances)/len(validation_performances)
performances.append(
(average_validation_performance,
average_training_performance,
f"model{model_counter}",
model_parameters_run)
)
model_counter += 1
torch.cuda.empty_cache()
sorted_performances = sorted(performances, key=lambda tup: (tup[0], tup[1]), reverse=True)
print(f"Best overall model was {sorted_performances[0][2]} with a validation F1 of {sorted_performances[0][0]} "
f"and a training F1 of {sorted_performances[0][1]}.")
print(f"Used parameters for best model:\n{sorted_performances[0][3]}")
return None
def make_predictions(model, data_loader, optimizer, loss_function, sigmoid,
prediction_list, label_list, backpropagate=True):
loss = 0
loss_count = 0
for data_batch, _ in data_loader:
if backpropagate:
optimizer.zero_grad()
data_batch = data_batch.to(select_device())
predictions = model.forward(
data_batch.x,
data_batch.edge_index,
data_batch.edge_index_cutoff,
data_batch.edge_features,
data_batch.dssp_features
)
loss_el = loss_function(predictions, data_batch.y)
# predictions is a tensor of shape: 69203, 3 num_nodes, data_graph.batch is tensor 69k, 2
loss_norm = torch.sum(loss_el)
loss += loss_norm.item()
loss_count += 1
predictions = sigmoid(predictions)
prediction_list.append(predictions.detach().cpu())
label_list.append(data_batch.y.detach().cpu())
if backpropagate:
loss_norm.backward()
optimizer.step()
torch.cuda.empty_cache()
return loss, loss_count
def train_epoch(model, train_loader, optimizer, loss_function, sigmoid):
prediction_list = []
label_list = []
model.train()
train_loss, train_loss_count = make_predictions(
model,
train_loader,
optimizer,
loss_function,
sigmoid,
prediction_list,
label_list,
backpropagate=True
)
return prediction_list, label_list, train_loss, train_loss_count
def validate_epoch(model, validation_loader, loss_function, sigmoid):
prediction_list = []
label_list = []
model.eval()
validation_loss, validation_loss_count = make_predictions(
model,
validation_loader,
None,
loss_function,
sigmoid,
prediction_list,
label_list,
backpropagate=False
)
return prediction_list, label_list, validation_loss, validation_loss_count
def train_and_validate(model, training_dataset, validation_dataset, model_parameters):
train_loader = DataLoader(
training_dataset,
batch_size=model_parameters["batchsize"],
shuffle=True,
pin_memory=True
)
validation_loader = DataLoader(
validation_dataset,
batch_size=model_parameters["batchsize"],
shuffle=True
)
training_evaluator = BindingResiduePredictionEvaluator()
validation_evaluator = BindingResiduePredictionEvaluator()
loss_function = torch.nn.BCEWithLogitsLoss(
reduction="none",
pos_weight=torch.tensor(model_parameters["weights"])
)
sigmoid = torch.nn.Sigmoid()
model.to(select_device())
optimizer = torch.optim.Adam(model.parameters(), **get_optimizer_arguments(only_first_value=True))
checkpoint_file = "checkpoint_file_early_stopping.pt"
early_stopping_patience = 10
early_stopping = EarlyStopping(
patience=early_stopping_patience, delta=0.01, checkpoint_file=checkpoint_file, verbose=True
)
epoch_counter = 0
for epoch in range(model_parameters["epochs"]):
if do_logging():
print("Epoch {}".format(epoch_counter))
epoch_counter += 1
torch.cuda.empty_cache()
# training
training_predictions, training_labels, train_loss, train_loss_count = train_epoch(
model,
train_loader,
optimizer,
loss_function,
sigmoid
)
# validation
validation_predictions, validation_labels, validation_loss, validation_loss_count = validate_epoch(
model,
validation_loader,
loss_function,
sigmoid
)
#evaluation
training_evaluator.evaluate_per_epoch(
training_predictions,
training_labels,
train_loss,
train_loss_count
)
validation_evaluator.evaluate_per_epoch(
validation_predictions,
validation_labels,
validation_loss,
validation_loss_count
)
if model_parameters["early_stopping"]:
f1_validation = validation_evaluator.performances["all"]["f1"][-1] * (-1)
# eval_val = val_loss
early_stopping(f1_validation, model, do_logging())
if early_stopping.early_stop:
training_evaluator.remove_last_x_performances(x=early_stopping_patience)
validation_evaluator.remove_last_x_performances(x=early_stopping_patience)
break
if model_parameters["early_stopping"]:
# load best model
model = torch.load(checkpoint_file)
optimizer.zero_grad(set_to_none=True)
return model, training_evaluator, validation_evaluator
def run_training(model_parameters, filename_model_prefix=""):
structure_cutoff = model_parameters["cutoff_structure"]
dataset = BindingResidueDatasetWithLabels(structure_cutoff)
cv_split_ids = get_cv_splits()
nr_of_splits = len(cv_split_ids)
training_evaluators = []
validation_evaluators = []
for i in range(nr_of_splits):
validation_ids = cv_split_ids[i]
training_ids = []
model = initialize_untrained_model(model_parameters)
for j in range(nr_of_splits):
if i != j:
training_ids += cv_split_ids[j]
training_dataset, validation_dataset = dataset.train_val_split(training_ids, validation_ids)
trained_model, training_evaluator, validation_evaluator \
= train_and_validate(model, training_dataset, validation_dataset, model_parameters)
# save model
if not Path(get_weight_dir()).exists():
Path(get_weight_dir()).mkdir(parents=True, exist_ok=True)
model_save_path = os.path.join(str(get_weight_dir()), f"{filename_model_prefix}trained_model_{i}.state_dict")
save_classifier_torch(trained_model, model_save_path)
# Write training details and final performance to file
training_evaluator.write_evaluation_results(f"{filename_model_prefix}model_training_{i}", model_parameters)
validation_evaluator.write_evaluation_results(f"{filename_model_prefix}model_validation_{i}", model_parameters)
training_evaluators.append(training_evaluator)
validation_evaluators.append(validation_evaluator)
return training_evaluators, validation_evaluators
|