def run_dae_cpd(df):
data = data_processing.load_dataset(df)
values = data.values
X, y = values[:, :-1], values[:, -1] # X : Samples, Y: Labels
## COMMENTED THIS OUT TO PREVENT A 20 MIN DELAY
#model = dae_cpd.dae(X, y).fit()
#result = model.fit_predict() # change point indexes
result = [
57, 211, 306, 359, 545, 691, 1176, 1319, 1412, 1820, 2277, 2470, 2559,
2696, 2872, 3179, 3270, 3357, 3564, 3645, 3772, 3976, 4148, 4287, 4349,
4408, 4472, 4557, 4758, 4856, 4933
]
print(result)
times = sorted(df['time'].unique())
print(len(times))
#print(times)
segments = [] # list of timestamp segments
segments.append((times[0], times[result[0]]))
for i in range(1, len(result)):
segments.append((times[result[i - 1]], times[result[i]]))
print(segments[:3])
return segments
def test():
# Chosen by Jordan as the best model
model_path = 'models/nc=62:F=3:M=5:lr=0.01:epoch=010.pt'
network = load_model(model_path)
full_dataset = load_dataset()
_, __, test = split_dataset(full_dataset)
test_acc = get_accuracy(network, test)
print(f'Test accuracy = {test_acc}')
def run(dataset_path):
model = Inception_V1(input_shape=(256, 256, 1))
#model = LeNet(input_shape = (256,256,1))
#model = ResNet50(input_shape = (256, 256, 1))
#model = Inception_small_V1(input_shape = (256,256,1), classes=2)
model.compile(optimizer='adam',
loss=[
'binary_crossentropy', 'binary_crossentropy',
'binary_crossentropy'
],
metrics=['accuracy'])
#model.compile(optimizer='adam', loss=['binary_crossentropy'], metrics=['accuracy'])
print("Loading dataset")
X_train, Y_train, X_valid, Y_valid = load_dataset(dataset_path)
#np.save('./X_train.npy', X_train)
#np.save('./Y_train.npy', Y_train)
#np.save('./X_valid.npy', X_valid)
#np.save('./Y_valid.npy', Y_valid)
#extract_masks('./segments', './ground_truth', './data')
#X_train = np.load('./X_train.npy')
#Y_train = np.load('./Y_train.npy')
#X_valid = np.load('./X_valid.npy')
#Y_valid = np.load('./Y_valid.npy')
#directories, dir_names = get_patient_directories(dataset_path)
#get_lung_masks(directories, dir_names)
#segment = np.load('./segment.npy')
#print(dataset.pixel_array[250])
Y_train = convert_to_one_hot(Y_train)
Y_valid = convert_to_one_hot(Y_valid)
class_weight = {0: 1., 1: 1.}
#model.fit(X_train, [Y_train, Y_train, Y_train], epochs = 1, batch_size = 32, class_weight=class_weight)
model.fit(X_train,
Y_train,
epochs=1,
batch_size=32,
class_weight=class_weight,
validation_split=0.2)
results = model.predict(X_valid)
output = []
for i in range(len(results)):
if (results[i] < 0.5):
output.append(0)
else:
output.append(1)
num_equal = 0
for i in range(len(output)):
if (output[i] == Y_valid[i].astype(int)):
num_equal += 1
print(num_equal / len(output))
import pandas as pd
import data_processing as dp
import plot_computations as pc
import plotting
import markdown
DEBUG = True
external_stylesheets = ['https://maxcdn.bootstrapcdn.com/bootstrap/4.0.0/css/bootstrap.min.css']
app = dash.Dash(__name__, external_stylesheets=external_stylesheets)
server = app.server
df = dp.load_dataset()
primary_energy_df = dp.load_primary_energy_sources(df)
with open(os.path.join("data", "united_states.geojson")) as infile:
united_states_geojson = json.load(infile)
# Precomputed figures
# us_main_plot_dict = pc.precompute_main_plots(df, primary_energy_df)
# state_total_dict = pc.precompute_state_per_year(df)
app.layout = html.Div(children = [
# html.Div(
# children=[
# html.Div(
# dcc.Graph(
# id="choropleth",
# figure=pc.update_choropleth(df, united_states_geojson)
def main():
network = CharacterClassifier()
full_dataset = load_dataset()
train, val, test = split_dataset(full_dataset, batch_size=64)
path = train_network(network, train, val, num_epochs=128)
plot_curves(path)
def verify_on_small_dataset():
network = CharacterClassifier(num_classes=4)
full_dataset = load_dataset()
dataloader = get_small_dataloader(full_dataset, num_classes=4)
path = train_network(network, dataloader)
plot_curves(path, val=False)