def experiment():
train_input_fn = generate_input_fn(is_train=True,
tfrecords_path=config.tfrecords_path,
batch_size=config.batch_size,
time_step=config.time_step)
eval_input_fn = generate_input_fn(is_train=False,
tfrecords_path=config.tfrecords_path,
batch_size=config.batch_size_eval,
time_step=config.time_step_eval)
estimator = Estimator(train_input_fn=train_input_fn,
eval_input_fn=eval_input_fn,
model_fn=model_fn)
estimator.train()
import numpy
# zde se importuje konkretni model
import models_source.model03 as source
create_model = source.create_model
epochs = source.epochs
batch_size = source.batch_size
# Priprav DATA
dataset = DatasetLoader("data/demography.csv", "data/Results_snemovna2017.csv")
inputset = dataset.get_input_acc_to_scheme(DataScheme.ALL)
outputset = dataset.get_outputset()
metadata = dataset.metadata
# test MODEL
estimator = Estimator(create_model)
estimator.train(inputset, outputset, epochs=epochs, batch_size=batch_size)
# jmeno python souboru jako string
modelname = (create_model.__module__).split('.')[1]
estimator.save_prediction_to_file(inputset, metadata,
'data/predictions/' + modelname + '.csv')
# uloz model do souboru .h5
estimator.save_model_to_file()
#numpy.save('trained/' + modelname + '-' + str(epochs) + 'epoch' +
# str(batch_size) + 'batch' + '.npy', result)
'type': Layers.FC,
'input_len': 30,
'num_neurons': 10,
'activation': Activations.SIGMOID
},
]
# layers = [
# {'type': Layers.INPUT, 'dimensions': (1, 28, 28)},
# {'type': Layers.CONV, 'stride': 1, 'fshape': (6, 1, 3, 3), 'padding': 1, 'activation': Activations.RELU},
# {'type': Layers.MAXPOOL, 'stride': 3, 'length': 3},
# {'type': Layers.CONV, 'stride': 1, 'fshape': (12, 6, 3, 3), 'padding': 1, 'activation': Activations.RELU},
# {'type': Layers.MAXPOOL, 'stride': 3, 'length': 3},
# {'type': Layers.CONV, 'stride': 1, 'fshape': (24, 12, 3, 3), 'padding': 1, 'activation': Activations.RELU},
# {'type': Layers.MAXPOOL, 'stride': 3, 'length': 3},
# {'type': Layers.CONV, 'stride': 1, 'fshape': (24, 24, 1, 1), 'padding': 0, 'activation': Activations.RELU},
# {'type': Layers.CONV, 'stride': 1, 'fshape': (10, 24, 1, 1), 'padding': 0, 'activation': Activations.RELU},
# {'type': Layers.FLATTEN, 'input_len': 24, 'num_neurons': 10, 'activation': Activations.SIGMOID},
# ]
network = Network(layers)
optimizer = MiniBatchGradientDescent(network,
learning_rate=0.01,
batch_size=30)
estimator = Estimator(network=network,
optimizer=optimizer,
dataset=mnist,
transformer=transform_data)
estimator.train(epochs=300)
estimator.test()