print max_length
###############
# Build Model #
###############
# Symbolic variables
x = T.tensor3(dtype=theano.config.floatX)
y = T.imatrix()
mask = T.ivector()
# Construct RNN class
classifier = RNN(input=x,
n_in=INPUT_DIM,
n_hidden=NEURONS_PER_LAYER,
n_out=OUTPUT_DIM,
n_layers=HIDDEN_LAYERS,
n_total=max_length,
batch=BATCH_SIZE,
mask=mask)
# Cost
cost = classifier.negative_log_likelihood(y) + classifier.L2_sqr
# Build Gradient
dparams = [T.grad(cost, param) for param in classifier.params]
# Build Train Model
print "Building Train Model..."
train_model = theano.function(inputs=[x, y, mask],
outputs=cost,
updates=Update(classifier.params, dparams,
###############
# Build Model #
###############
# Symbolic variables
x = T.tensor3(dtype=theano.config.floatX)
y = T.imatrix()
mask = T.ivector()
# Construct RNN class
classifier = RNN(
input=x,
n_in=INPUT_DIM,
n_hidden=NEURONS_PER_LAYER,
n_out=OUTPUT_DIM,
n_layers=HIDDEN_LAYERS,
n_total=max_length,
batch=BATCH_SIZE,
mask=mask,
)
# Cost
cost = classifier.negative_log_likelihood(y) + classifier.L2_sqr
# Build Gradient
dparams = [T.grad(cost, param) for param in classifier.params]
# Build Train Model
print "Building Train Model..."
train_model = theano.function(
inputs=[x, y, mask], outputs=cost, updates=Update(classifier.params, dparams, classifier.velo)
def get_model(self, input_size, hidden_size, output_size) -> NeuralNetwork:
return RNN(input_size, hidden_size, 12, output_size)
observation, reward, done, _ = env.step(action)
if done:
break
if __name__ == '__main__':
env = gym.make('BipedalWalker-v2')
env = gym.wrappers.Monitor(env, 'bipedalwalker', video_callable=lambda episode_id: True, force=True)
env.seed(123)
observation = env.reset()
INPUT_SIZE = 24
HIDDEN_SIZE = 16
OUTPUT_SIZE = 4
rnn = RNN(10, 24, 12, OUTPUT_SIZE)
# rnn.load('../../../models/bipedalwalker/09-23-2019_07-09_NN=RNNIndividual_POPSIZE=50_GEN=2000_PMUTATION_0'
# '.3_PCROSSOVER_0.8.npy')
# test_rnn()
mlp = MLP(INPUT_SIZE, HIDDEN_SIZE, OUTPUT_SIZE)
# mlp.load("../../../models/bipedalwalker/09-07-2019_16-34-56_POPSIZE=30_GEN=5_MUTATION_0.609-07-2019_16-34"
# "-56_POPSIZE=30_GEN=5_MUTATION_0.6.npy")
# test_mlp()
# Model 09-14-2019 NN: 24 - 32 - 12 - 4
# Model 09-21-2019 NN: 10 - 24 - 12 - 4
# Model 09-26-2019 NN: 5 - 16 - 12 - 4
# Model 09-30-2019 NN: 10 - 16 - 12 - 4
mlp_torch = MLPTorch(10, 16, 12, OUTPUT_SIZE)
mlp_torch.load("../../../models/bipedalwalker/10-21-2019_02-57_NN=MLPTorchIndividual_POPSIZE=30_GEN"
test_num1 = len(test_index1) - 1
test_num2 = len(test_index2) - 1
print test_num1
print test_num2
# Symbolic variables
x = T.tensor3(dtype=theano.config.floatX)
y = T.imatrix()
mask = T.ivector()
# Construct RNN class
classifier = RNN(
input=x,
n_in=INPUT_DIM,
n_hidden=NEURONS_PER_LAYER,
n_out=OUTPUT_DIM,
n_layers=HIDDEN_LAYERS,
n_total=max_length,
batch=BATCH_SIZE,
mask=mask
)
classifier.load_model(args.model_in)
# Build Test Model
print "Building Test Model"
test_model = theano.function(
inputs=[x,mask],
outputs=classifier.y_pred
)
# Create Phone Map
