def train(args):
mlp = MLPClassifier(hidden_layer_sizes=args.layers,
alpha=args.alpha,
learning_rate_init=args.learning_rate,
max_iter=args.epochs)
models.train_model(mlp, args.feature_extractor, args.num_examples)
return mlp
def parapharse_models(glove, train, dev, ppdb_file):
ppdb = paraphrase.load_parap(ppdb_file)
aug = augmented_dataset(glove, train, ppdb)
train_aug = train + aug
models.train_model(train_aug, dev, glove, model_filename = 'models/train_aug')
models.train_model(train, dev, glove, model_filename = 'models/train_noaug')
def mixture_experiments(train, dev, glove, splits = 5):
for i in range(splits):
model_name = 'mixture' + str(i)
print 'Model', model_name
model = models.init_model()
div = len(train) / splits
models.train_model(train[:i*div] + train[(i+1)*div:splits*div], dev, glove, model, 'models/' + model_name)
def construct_model(model_paths):
models.train_model(model_paths, transfer=True)
models.calculate_predictions(model_paths.test_solutions,
model_paths.test_image_path,
model_paths.test_true, model_paths.test_preds,
model_paths.checkpoint_overall_path)
models.eval_metrics(model_paths.test_true, model_paths.test_preds,
model_paths.test_conf_matrix,
model_paths.test_other_metrics)
def grid_experiments(train, dev, glove, embed_size = 300, hidden_size = 100):
lr_vec = [0.001, 0.0003, 0.0001]
dropout_vec = [0.0, 0.1, 0.2]
reg_vec = [0.0, 0.001, 0.0003, 0.0001]
for params in itertools.product(lr_vec, dropout_vec, reg_vec):
filename = 'lr' + str(params[0]).replace('.','') + '_drop' + str(params[1]).replace('.','') + '_reg' + str(params[2]).replace('.','')
print 'Model', filename
model = models.init_model(embed_size, hidden_size, params[0], params[1], params[2])
models.train_model(train, dev, glove, model, 'models/' + filename)
def extended_tautologies(train, dev, glove):
augment_data = generate_all(train)
from random import shuffle
shuffle(augment_data)
augment_weight = [0, 0.05, 0.15, 0.5]
for w in augment_weight:
new_train = train + augment_data[:int(len(train)*w)]
str = str(w).replace('.','')
model = models.init_model()
models.train_model(new_train, dev, glove, model = model, model_dir = 'models/aug' + w_str)
def show_spiral_demo(baseline_model_weights_file=None,
mixup_model_weights_file=None):
print("~~~~~~~~~~~~~~~~~")
print("~~ SPIRAL DEMO ~~")
print("~~~~~~~~~~~~~~~~~")
x_train, y_train, x_val, y_val, x_test, y_test = data_loader.get_two_spirals_data(
2000, noise=0.9)
# visualizer.plot_spiral_dataset(x_train, y_train)
spiral_model = models.create_spiral_model(models.MixupMode.NO_MIXUP)
spiral_model_with_mixup = models.create_spiral_model(
models.MixupMode.MANIFOLD_MIXUP)
print("baseline spiral model")
if baseline_model_weights_file:
print("Loading model from file {}...".format(
baseline_model_weights_file))
spiral_model.load_weights(baseline_model_weights_file)
else:
print("Training...")
models.train_model(
spiral_model, (x_train, y_train), (x_val, y_val),
batch_size=20,
save_to_file=get_path_for_trained_models(spiral_model))
print("Test Accuracy: {:.3f}".format(
spiral_model.get_accuracy(x_test, y_test)))
print("Spiral model with manifold mixup")
if mixup_model_weights_file:
print("Loading model from file {}...".format(mixup_model_weights_file))
spiral_model_with_mixup.load_weights(mixup_model_weights_file)
else:
print("Training...")
models.train_model(
spiral_model_with_mixup, (x_train, y_train), (x_val, y_val),
batch_size=20,
save_to_file=get_path_for_trained_models(spiral_model_with_mixup))
print("Test Accuracy: {:.3f}".format(
spiral_model_with_mixup.get_accuracy(x_test, y_test)))
visualizer.plot_spiral_model_confidence(spiral_model,
x_train,
y_train,
title='no mixup')
visualizer.plot_spiral_model_confidence(spiral_model_with_mixup,
x_train,
y_train,
title='manifold mixup')
def main():
add_transformed_vars = True
all_predictions = None
for label_column in args.label_columns:
print('Training for Label {}'.format(label_column))
(all_features_predictions_df, train_features, transformed_train_labels,
train_labels, test_features, transformed_test_labels, test_labels,
all_feature_names, all_label_names) = utils.prepare_data(
args.input_features_filename, args.input_labels_filename,
args.input_predictions_filename, args.transformation,
label_column, args.train_size, args.test_size,
add_transformed_vars, False, False)
all_features = np.concatenate((train_features, test_features), axis=0)
all_labels = np.concatenate((train_labels, test_labels), axis=0)
all_transformed_labels = np.concatenate(
(transformed_train_labels, transformed_test_labels), axis=0)
#print 'Max label values: {}'.format(np.max(all_labels))
#print 'Min label values: {}'.format(np.min(all_labels))
if all_predictions is None:
all_predictions = all_features_predictions_df[all_label_names]
#print('\n*****************************\n')
(model, best_params) = models.train_model(
all_features, all_transformed_labels, args.algorithm,
args.num_alphas, args.skip_cross_validation, args.alpha,
args.l1_ratio, args.num_jobs)
train_y_true = all_labels
train_y_pred = utils.predict(all_features, model, label_column)
#print 'R-Squared on train is {}'.format(r2_score(train_y_true, train_y_pred))
#print('\n')
#print 'MAE on train: {}'.format(mean_absolute_error(train_y_true, train_y_pred))
#print 'MSE on train: {}'.format(mean_squared_error(train_y_true, train_y_pred))
coefs = model.coef_
print('Optimal parameters in final model is {}'.format(best_params))
# info on most predictive coefs
print('Model has {} non-zero coefficients.'.format(
len(coefs[coefs != 0])))
highest_indices = np.abs(coefs).argsort()[::-1]
#print 'Top predictors of {}'.format(label_column)
for i in range(min(10, len(coefs[coefs != 0]))):
idx = highest_indices[i]
# print '{}: {}'.format(all_feature_names[idx], coefs[idx])
# predict on all and write to file
prediction_features = all_features_predictions_df[all_feature_names]
label_predictions = utils.predict(prediction_features, model,
label_column)
# correct if below/above min/max
label_predictions[label_predictions < np.min(all_labels)] = np.min(
all_labels)
label_predictions[label_predictions > np.max(all_labels)] = np.max(
all_labels)
all_predictions.loc[:, label_column] = label_predictions
print('\n*****************************')
print('*****************************\n')
all_predictions.to_csv(args.output_prediction_filename)
def run(self):
w = self.img_width
h = self.img_height
input_shape = (h, w, 1)
tuned_path = 'tuned_model.h5'
original_model = 'classification_model.h5'
if not os.path.isfile(tuned_path):
shutil.copyfile(original_model, tuned_path)
builder = build_classification_model(input_shape=input_shape,
num_classes=self.NUM_CLASSES)
builder.load_weights(tuned_path)
classifier = builder.get_complete_model(input_shape=input_shape)
while True:
examples, epochs = self.get_job()
total = len(examples)
m_train = int(round(total * self.SPLIT_RATIO))
m_val = total - m_train
shuffle(examples)
train_examples = examples[:m_train]
val_examples = examples[m_train:]
train_gen = self.get_generator(train_examples, self.BATCH_SIZE)
val_gen = self.get_generator(val_examples, self.BATCH_SIZE)
train_model(model=classifier,
train_gen=train_gen,
validation_gen=val_gen,
m_train=m_train,
m_val=m_val,
mini_batch_size=self.BATCH_SIZE,
save_path=tuned_path,
epochs=epochs)
metrics = classifier.evaluate_generator(
generator=self.get_generator(val_examples, self.BATCH_SIZE),
steps=calculate_num_steps(m_val, self.BATCH_SIZE))
self.completed.emit(metrics[-1])
def show_bottleneck_representation_demo(baseline_weights_file=None,
manifold_mixup_weights_file=None):
print("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~")
print("~~ BOTTLENECK REPRESENTATION DEMO ~~")
print("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~")
x_train, y_train, x_val, y_val, x_test, y_test = data_loader.get_mnist_data(
)
mnist_model_b2 = models.create_mnist_model_bottleneck_2(
mixup_mode=MixupMode.NO_MIXUP)
mnist_model_b2_with_mixup = models.create_mnist_model_bottleneck_2(
mixup_mode=MixupMode.MANIFOLD_MIXUP)
print("baseline mnist 2-node bottelneck model")
if baseline_weights_file:
print("Loading model from file {}...".format(baseline_weights_file))
mnist_model_b2.load_weights(baseline_weights_file)
else:
print("Training...")
models.train_model(
mnist_model_b2, (x_train, y_train), (x_val, y_val),
save_to_file=get_path_for_trained_models(mnist_model_b2))
print("Test Accuracy: {:.3f}".format(
mnist_model_b2.get_accuracy(x_test, y_test)))
print("mnist 2-node bottleneck model with manifold mixup")
if manifold_mixup_weights_file:
print("Loading model from file {}...".format(
manifold_mixup_weights_file))
mnist_model_b2_with_mixup.load_weights(manifold_mixup_weights_file)
else:
print("Training...")
models.train_model(mnist_model_b2_with_mixup, (x_train, y_train),
(x_val, y_val),
save_to_file=get_path_for_trained_models(
mnist_model_b2_with_mixup))
print("Test Accuracy: {:.3f}".format(
mnist_model_b2_with_mixup.get_accuracy(x_test, y_test)))
visualizer.show_b2_model_hidden_representation(mnist_model_b2, x_train,
y_train)
visualizer.show_b2_model_hidden_representation(mnist_model_b2_with_mixup,
x_train, y_train)
def main():
if len(sys.argv) <= 2:
print("Please provide the inputfile(realtimedata)", len(sys.argv))
exit(-1)
window = 20
window2 = 10
user_rating = 5
chances = 0.2
inputfile = sys.argv[1]
out_dir = sys.argv[1].split('.')[0] + "_filtered"
if len(sys.argv) == 3:
print("User past rating out of 10", int(sys.argv[2]))
user_rating = int(sys.argv[2])
baseline_dataset = load_my_data("basedata/main.csv", ",")
baseline_dataset = preprocess_data(baseline_dataset, window, window2)
#print(baseline_dataset)
m_mdls = mdl.train_model(baseline_dataset)
print("DONE TRAINING MODELS WITH BASELINE DATASET")
#os.mkdir(out_dir)
filename = inputfile
dataset1 = load_my_data(inputfile, ',')
sliding_samples = 120
rows = dataset1.shape[0]
for timestep in range(int(rows / sliding_samples) * sliding_samples):
start = timestep
end = timestep + sliding_samples
print("PROCESSING WINDOW FROM ", start, " UNTIL ", end)
if end > rows:
end = rows - 1
if end - start + 1 < 12:
timestep += sliding_samples
continue
dset1 = dataset1[start:end]
dset1 = preprocess_data(dset1, window, window2)
#print(dset1)
out_lbls, prob = mdl.get_predictions(dset1, m_mdls)
chances, user_rating = get_accident_probability(
user_rating, out_lbls, prob, chances)
#write_to_file(out_dir, inputfile.split('.')[0], dataset1)
timestep += sliding_samples
print("CHANCES OF ACCIDENT:", chances, " NEW USER RATING:",
round(user_rating))
def train_pipeline(training_pipeline_params: TrainingPipelineParams):
logger.info(f"Start training with params: {training_pipeline_params}")
load_data(training_pipeline_params.input_data_path,
training_pipeline_params.input_data_url)
data = read_data(training_pipeline_params.input_data_path)
logger.info(f"Raw data shape: {data.shape}")
train_df, val_df = split_train_val_data(
data, training_pipeline_params.splitting_params)
logger.info(f"Train df shape: {train_df.shape}")
logger.info(f"Val df shape: {val_df.shape}")
pipeline = build_transformer(training_pipeline_params.feature_params)
pipeline.fit(train_df)
logger.info(f"Transform fitted.")
train_features = make_features(pipeline, train_df)
train_target = extract_target(train_df,
training_pipeline_params.feature_params)
logger.info(f"Train features shape: {train_features.shape}")
val_features = make_features(pipeline, val_df)
val_target = extract_target(val_df,
training_pipeline_params.feature_params)
logger.info(f"Val features shape: {train_features.shape}")
model = get_model(training_pipeline_params.train_params)
model = train_model(train_features, train_target, model)
logger.info(f"Model trained.")
predictions = predict_model(val_features, model)
metrics = evaluate_model(predictions, val_target)
path_to_model = save_artifacts(metrics, model, pipeline,
training_pipeline_params)
return path_to_model, metrics
###############################################################################
# Creating and training the model #
###############################################################################
pl.seed_everything(3258) # for reproducibility
model = ViscNet(hparams)
feats = model.composition_to_features(composition_df)
model.update_x_norm(feats.loc[train_val_idx].values)
train_ds, val_ds = train_val_ds(feats, viscosity_df, train_idx, val_idx)
train_model(
model,
train_ds,
val_ds,
num_workers=4,
deterministic=True,
)
torch.save(model, r'files/viscnet.pt')
state_dict = model.state_dict()
# Learning curve
lc_train = model.learning_curve_train
lc_val = model.learning_curve_val[:-1]
lc_x = np.arange(1, len(lc_train) + 1)
learning_curve(lc_x, lc_train, lc_val, [1, len(lc_train)], 'ViscNet')
###############################################################################
def main():
print('Initializing...')
base_path = os.path.dirname(sys.argv[0])
images_paths = [
os.path.normpath(os.path.join(base_path, 'datasets/', wwu_muenster_images_path)),
os.path.normpath(os.path.join(base_path, 'datasets/', arte_lab_images_path))
]
masks_paths = [
os.path.normpath(os.path.join(base_path, 'datasets/', wwu_muenster_masks_path)),
os.path.normpath(os.path.join(base_path, 'datasets/', arte_lab_masks_path))
]
models_path = os.path.normpath(os.path.join(base_path, 'models/'))
# Load data
images = None
masks = None
for images_path, masks_path in zip(images_paths, masks_paths):
images = storage.load_datamap(images_path, datamap=images)
masks = storage.load_datamap(masks_path, datamap=masks)
# Create dataset
dataset = datasets.create_dataset(images, masks, rescale=(256, 256))
# Augment dataset
dataset = datasets.augment_dataset(dataset, new_size=2500)
# Create model
model = models.create_model()
model.summary()
loss_function = {
'mse': tf.keras.losses.MeanSquaredError(name='mse'),
'mae': tf.keras.losses.MeanAbsoluteError(name='mae'),
'bce': tf.keras.losses.BinaryCrossentropy(name='bce')
}[sys.argv[1]] if len(sys.argv) > 1 else tf.keras.losses.MeanSquaredError(name='mse')
epochs = int(sys.argv[2]) if len(sys.argv) > 2 else 20
print(loss_function)
# Load weights
if utils.input_boolean('Load weights?'):
try:
model.load_weights(os.path.join(models_path, loss_function.name))
except Exception as ex:
print('Load error!')
print(ex)
# Predict images (pre-training)
print('Showing pre-training example predictions...')
for index in range(5):
predictions = model.predict(np.array([dataset.images[index]]))
utils.print_prediction(dataset.images[index], dataset.labels[index], predictions[0])
# Train model
models.train_model(
dataset, model,
loss_function=loss_function,
epochs=epochs
)
# Predict images (post-training)
print('Showing post-training example predictions...')
for index in range(5):
predictions = model.predict(np.array([dataset.images[index]]))
utils.print_prediction(dataset.images[index], dataset.labels[index], predictions[0])
# Save weights
if utils.input_boolean('Save weights?'):
try:
model.save_weights(os.path.join(models_path, loss_function.name))
except Exception as ex:
print('Save error!')
print(ex)
print('Terminating...')
return model
def train(args):
svc = SVC(C=args.regularization, kernel=args.kernel, gamma=args.gamma)
models.train_model(svc, args.feature_extractor, args.num_examples)
return svc
from models import lstm_model
from models import train_model
from preprocess import load_data
from preprocess import load_fast_text_embedding
from preprocess import get_embed_weights
max_len = 50
embedding_dim = 300
tokenizer, x_train, y_train, x_test, y_test, vocab_size = load_data(
'sarcasm_v2.csv')
embedding_index = load_fast_text_embedding('wiki-news-300d-1M-subword.vec')
embedding_matrix = get_embed_weights(embedding_index, tokenizer)
model = lstm_model(vocab_size, embedding_matrix)
train_model(model, x_train, y_train, x_test, y_test)
#half embeddingd
# Validation Loss:1.128701367992565 Validation Accuracy:0.6840490793889286
# Validation Accuracy:68.40% (+/- 0.00%)
# Train Loss:1.1293133928731907 Train Accuracy:68.40490797546013
# Test Loss:1.110706667958593 Test Accuracy:71.16564420834641
dataset = dataset.filter(
lambda t, y, s: tf.equal(tf.shape(y)[0], batch_size))
iterator = dataset.make_initializable_iterator()
return iterator
batch_size = 32
num_epochs = 200
restore = True
filename = tf.placeholder(tf.string, shape=[])
iterator = create_iterator(filename, batch_size)
length, token, label = iterator.get_next()
output = train_model(token, label, length, batch_size)
infer_output = inference_model(token, label, length, batch_size)
pred = tf.argmax(output, axis=2)
loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=label,
logits=output)
cost = tf.reduce_mean(loss)
updates = tf.train.AdamOptimizer(1e-4).minimize(cost)
sess = tf.Session()
saver = tf.train.Saver()
sess.run(tf.global_variables_initializer())
if restore == True:
saver.restore(sess, tf.train.latest_checkpoint('../models'))
def run(args):
test_loss = models.train_model(args, ex)
ex.log_scalar('loss', test_loss)
return test_loss
54,
3,
feature_extract=feat_extract)
elif model_type == "S":
model = mdl.StateModel(54, 3)
# create loss function
criterion = nn.MSELoss(reduction='sum')
# define optimization method
optimizer = opt.Adam(model.parameters(), lr=lr)
model, train_history, val_history, val_loss = mdl.train_model(
model,
criterion,
optimizer,
dataloaders,
dataset_sizes,
num_epochs=50,
model_type=model_type,
weight_file=weight_file,
suppress_log=False,
hyperparam_search=True)
if val_loss < best_loss:
print('found better lr: {}, loss:{}'.format(lr, val_loss))
print('previous best lr:{}, loss:{}'.format(best_lr, best_loss))
best_loss = val_loss
best_lr = lr
print("Learning Rate search completed, best LR:{}".format(best_lr))
#%%
weight_decay_list = [0, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5, 1e-6, 1e-7]
def detect_boxes():
aws.detect_boxes()
if __name__ == '__main__':
model_paths = models.initialize_default_paths()
system_arguments = ' '.join(sys.argv[1:])
if system_arguments == "aws":
detect_boxes()
if system_arguments == "Manip Data":
preprocess_data(model_paths)
elif system_arguments == "Train Model":
models.train_model(model_paths)
elif system_arguments == "Train Transfer Model":
models.train_model(model_paths, transfer=True)
elif system_arguments == "Predict":
models.calculate_predictions(model_paths.valid_solutions,
model_paths.valid_image_path,
model_paths.valid_true,
model_paths.valid_preds,
model_paths.checkpoint_overall_path)
models.eval_metrics(model_paths.valid_true, model_paths.valid_preds,
model_paths.valid_conf_matrix,
model_paths.valid_other_metrics)
elif system_arguments == "Predict Test":
def run_ablations(model, num_ablations):
'''set up some persistent tracking variables'''
remove_features = [] # list of features we are removing
metrics_list = [] # list storing dictionary of performance metrics
# feature indexes
full_state_index = np.arange(7, 61)
input_state = 54
# create loss function
criterion = nn.MSELoss(reduction='sum')
# define optimization method
optimizer = opt.Adam(model.parameters(), lr=0.01)
param_count = []
param_count.append(count_params(model))
current_feature_list = np.array(qty)
# create the dataloader
dataloaders, dataset_sizes = dat.init_dataset(train_list, val_list,
val_list, model_type,
config_dict)
print('evaluating full model predictions...')
predictions = mdl.evaluate_model(model,
dataloaders['test'],
model_type=model_type,
no_pbar=True)
# compute the loss statistics
print('computing full model performance metrics...')
metrics = model_eval.compute_loss_metrics(
predictions, dataloaders['test'].dataset.label_array[:, 1:4])
metrics_list.append(metrics)
print('Performance Summary of Full Model:')
print(metrics)
print('Running ablation study on model type:' + model_type)
for iteration in range(num_ablations):
print('-' * 10)
print('Begin ablation run: {}/{}'.format(iteration + 1, num_ablations))
print('-' * 10)
# compute the backprop values:
gbp_data = model_eval.compute_GBP(model,
dataloaders['test'],
num_state_inputs=input_state,
model_type=model_type,
no_pbar=True)
# evaluate means
df_gbp_means = model_eval.compute_and_plot_gbp(gbp_data,
current_feature_list,
True,
suppress_plots=True)
# group by feature type and rank by value
df_gbp_means = df_gbp_means.groupby('feature').mean().sort_values(
by='gbp', ascending=False).reset_index()
# get top ranking value and append to removal list
feature_to_remove = df_gbp_means.iloc[0, 0]
print("removing " + feature_to_remove + "...")
remove_features.append(feature_to_remove)
# create the mask
mask = np.isin(qty, remove_features, invert=True)
# mask the full state vector in config_dict global variable
config_dict['custom_state'] = full_state_index[mask]
current_feature_list = np.array(qty)[
mask] #update the current feature list
# decrease the input dimension of the model by one
input_state = input_state - 1
# redefine the models
print('redefining model with input state dims: {}'.format(input_state))
if model_type == "VS":
model = mdl.StateVisionModel(30,
input_state,
3,
feature_extract=feat_extract)
elif model_type == "S":
model = mdl.StateModel(input_state, 3)
# recalculate the number of parameters
param_count.append(count_params(model))
# redefine the optimizer
optimizer = opt.Adam(model.parameters(), lr=0.01)
# redefine the dataloader
dataloaders, dataset_sizes = dat.init_dataset(train_list, val_list,
val_list, model_type,
config_dict)
# retrain the model
model, train_history, val_history = mdl.train_model(
model,
criterion,
optimizer,
dataloaders,
dataset_sizes,
num_epochs=50,
model_type=model_type,
weight_file=weight_file,
no_pbar=True)
print('retraining completed')
# do inference
print('evaluating model predictions...')
predictions = mdl.evaluate_model(model,
dataloaders['test'],
model_type=model_type,
no_pbar=True)
# compute the loss statistics
print('computing performance metrics...')
metrics = model_eval.compute_loss_metrics(
predictions, dataloaders['test'].dataset.label_array[:, 1:4])
metrics_list.append(metrics)
print('Performance Summary:')
print(metrics)
return remove_features, param_count, metrics_list
# Add this project to the path
import os; import sys; currDir = os.path.dirname(os.path.realpath("__file__"))
rootDir = os.path.abspath(os.path.join(currDir, '..')); sys.path.insert(1, rootDir)
import warnings
warnings.filterwarnings("ignore")
# My modules
from models.train_model import *
from features.build_features import *
features_pipeline = data_preparation()
train_model(features_pipeline)
gs = ...
cross_val_score(estimator=gs
import numpy as np
import constants as cst
import import_data
import models
import attack_function
import visualization
if __name__ == '__main__':
# Load CIFAR10 data
x_train, y_train, x_test, y_test = import_data.format_data()
# Train or Load standard model (trained on not attacked data)
if cst.TRAIN_standard_model:
print("Training a new standard model")
model = models.train_model(x_train, y_train, cst.config_standard_model,
'standard_model')
else:
print("Loading standard model")
model = load_model(cst.STANDARD_trained_model)
# Attack Data
if cst.MAKE_ATTACK:
print("Attacking data")
x_test_attacked = attack_function.make_attack(
x_test, y_test, model, 'x_test_attacked_FGSM_0_03',
cst.attack_style)
else:
print("Load attacked data")
x_test_attacked = np.load(cst.ATTACKED_TEST)
if cst.VIZ:
mlp_layers_number=mlp_layers_number,
time_steps=seq_length,
learning_rate=learning_rate,
optimizer=optimizer,
metrics=metrics,
features_number=features_number)
train_model(model,
features_train,
annot_train,
features_valid,
annot_valid,
output_class_weights=[classWeights],
batch_size=batch_size,
epochs=epochs,
seq_length=seq_length,
save=save,
saveMonitor=saveMonitor,
saveMonitorMode=saveMonitorMode,
saveBestName=saveBestName,
reduceLrOnPlateau=reduceLrOnPlateau,
reduceLrMonitor=reduceLrMonitor,
reduceLrMonitorMode=reduceLrMonitorMode,
reduceLrPatience=reduceLrPatience,
reduceLrFactor=reduceLrFactor)
# Test
model.load_weights(saveBestName + '-best.hdf5')
#predict_test = np.zeros((annot_test.shape[1],annot_test.shape[2]))
nRound = annot_test.shape[1] // seq_length
timestepsRound = nRound * seq_length
def show_svd_demo(baseline_weights_file=None,
input_mixup_weights_file=None,
manifold_mixup_weights_file=None):
print("~~~~~~~~~~~~~~")
print("~~ SVD DEMO ~~")
print("~~~~~~~~~~~~~~")
x_train, y_train, x_val, y_val, x_test, y_test = data_loader.get_mnist_data(
)
mnist_model_b12 = models.create_mnist_model_bottleneck_12(
mixup_mode=MixupMode.NO_MIXUP)
mnist_model_b12_with_input_mixup = models.create_mnist_model_bottleneck_12(
mixup_mode=MixupMode.INPUT_MIXUP)
mnist_model_b12_with_manifold_mixup = models.create_mnist_model_bottleneck_12(
mixup_mode=MixupMode.MANIFOLD_MIXUP)
print("baseline mnist 12-node bottleneck model")
if baseline_weights_file:
print("Loading model from file {}...".format(baseline_weights_file))
mnist_model_b12.load_weights(baseline_weights_file)
else:
print("Training...")
models.train_model(
mnist_model_b12, (x_train, y_train), (x_val, y_val),
epochs=6,
save_to_file=get_path_for_trained_models(mnist_model_b12))
print("Test Accuracy: {:.3f}".format(
mnist_model_b12.get_accuracy(x_test, y_test)))
print("mnist 12-node bottleneck model with input mixup")
if input_mixup_weights_file:
print("Loading model from file {}...".format(input_mixup_weights_file))
mnist_model_b12_with_input_mixup.load_weights(input_mixup_weights_file)
else:
print("Training...")
models.train_model(mnist_model_b12_with_input_mixup,
(x_train, y_train), (x_val, y_val),
save_to_file=get_path_for_trained_models(
mnist_model_b12_with_input_mixup))
print("Test Accuracy: {:.3f}".format(
mnist_model_b12_with_input_mixup.get_accuracy(x_test, y_test)))
print("mnist 12-node bottleneck model with manifold mixup")
if manifold_mixup_weights_file:
print("Loading model from file {}...".format(
manifold_mixup_weights_file))
mnist_model_b12_with_manifold_mixup.load_weights(
manifold_mixup_weights_file)
else:
print("Training...")
models.train_model(mnist_model_b12_with_manifold_mixup,
(x_train, y_train), (x_val, y_val),
save_to_file=get_path_for_trained_models(
mnist_model_b12_with_manifold_mixup))
print("Test Accuracy: {:.3f}".format(
mnist_model_b12_with_manifold_mixup.get_accuracy(x_test, y_test)))
visualizer.compare_svd_for_b12_models([
mnist_model_b12, mnist_model_b12_with_input_mixup,
mnist_model_b12_with_manifold_mixup
], x_train, y_train)
fig_save_path = Path(r'./plots/')
# path for the model trained with a reserved holdout dataset
h_path = Path(rf'./model_files/experiment_{exp_num:02d}_model_with_holdout.pt')
# path for the model trained without a reserved holdout dataset
f_path = Path(rf'./model_files/experiment_{exp_num:02d}_model_final.pt')
### Code
data = get_data(compounds, round_comp_decimal=3, round_temperature_decimal=0)
model_h = train_model(Model,
patience,
data,
compounds,
holdout_size,
dataloader_num_workers,
max_epochs,
save_path=h_path)
model_f = train_model(Model,
patience,
data,
compounds,
False,
dataloader_num_workers,
max_epochs,
save_path=f_path)
if COMPUTE_METRICS:
from functools import partial
# Data and model checkpoints directories
parser.add_argument('--data_dir', type=str, default='',
help='data directory containing input.txt with training examples')
args = parser.parse_args()
files = os.listdir(args.data_dir)
print(files)
##
# Need to save model weights, history to S3
##
# Save final model out to opt.
##
files = ['ID_0a336e630', 'ID_0ba79c0ef', 'ID_0bc7199c6']
#path = '../Example Bucket'
train_gen = dw.DataGenerator(folder=args.data_dir,batch_size=1, file_list=files, shuffle=False)
test_gen = dw.DataGenerator(folder=args.data_dir,batch_size=1, file_list=files, shuffle=False)
model = md.unet2D(input_size = (512,512,4))
history = md.train_model(model, train_gen, test_gen, name="model", checkpoint_dir=args.final_model, epochs=3)
print(args.final_model)
#model.save(args.final_model + '/trainedmodel.h5') # saving the model
with open(args.final_model + '/trainHistoryOld', 'wb') as handle: # saving the history of the model
pickle.dump(history.history, handle)
model = mdl.StateVisionModel(30,
54,
3,
feature_extract=feat_extract,
TFN=True)
elif model_type == "S":
model = mdl.StateModel(54, 3)
elif (model_type == "V") or (model_type == "V_RNN"):
#model = mdl.VisionModel(3)
model = mdl.BabyVisionModel()
weight_file = weight_file + "_fffsdata.dat"
# create loss function
criterion = nn.MSELoss(reduction='sum')
# define optimization method
optimizer = opt.Adam(model.parameters(), lr=1e-3, weight_decay=0)
#optimizer = opt.SGD(model.parameters(),lr=1e-5,weight_decay=0,momentum=0.9)
model, train_history, val_history, _ = mdl.train_model(
model,
criterion,
optimizer,
dataloaders,
dataset_sizes,
num_epochs=100,
L1_loss=1e-3,
model_type=model_type,
weight_file=weight_file,
suppress_log=False,
multigpu=False)
img_height=imgHeight,
cnnType=cnnType,
cnnFirstTrainedLayer=cnnFirstTrainedLayer,
cnnReduceDim=cnnReduceDim)
history = train_model(model,
features_train,
annot_train,
features_valid,
annot_valid,
output_class_weights=[classWeightFinal],
batch_size=batch_size,
epochs=epochs,
seq_length=seq_length,
save=save,
saveMonitor=saveMonitor,
saveMonitorMode=saveMonitorMode,
saveBestName=saveBestName,
reduceLrOnPlateau=reduceLrOnPlateau,
reduceLrMonitor=reduceLrMonitor,
reduceLrMonitorMode=reduceLrMonitorMode,
reduceLrPatience=reduceLrPatience,
reduceLrFactor=reduceLrFactor,
features_type=inputFeaturesFrames,
img_width=imgWidth,
img_height=imgHeight,
cnnType=cnnType)
# Results
print('Results')
model.load_weights(saveBestName + '-best.hdf5')
dataGlobal[outputName][timeString]['results'] = {}
import sys
from lib import upload_dataset
from models import train_model, measure_test_accuracy
architecture = int(sys.argv[1])
batch_size = int(sys.argv[2])
latent_dim = int(sys.argv[3])
epochs = int(sys.argv[4])
print("yes 1")
(train_encoder_input, train_decoder_input, train_decoder_target), \
(test_encoder_input, test_decoder_input, test_decoder_target) = upload_dataset()
print(train_encoder_input.shape, train_decoder_input.shape, train_decoder_target.shape)
model, encoder_states = train_model(encoder_input_data=train_encoder_input,
decoder_input_data=train_decoder_input,
decoder_target_data=train_decoder_target,
batch_size=batch_size,
latent_dim=latent_dim,
epochs=epochs,
model_architecture=architecture)
# accuracy = measure_test_accuracy(test_decoder_input, model, encoder_states, latent_dim=512)