def hyperopt(nofm_layer1, fshape_layer1, learning_rate_layer1, coef_layer1,
fshape_layer2, stride_layer2, nofm_layer3, fshape_layer3,
learning_rate_layer3, coef_layer3, fshape_layer4, stride_layer4,
learning_rate_layer5, coef_layer5, learning_rate_output, coef_output,
num_epochs, batch_size):
basepath = "/Users/DOE6903584/NERSC/mantissa-new/AR/data"
repo_path = os.path.join(basepath, "/results/")
fland = os.path.join(basepath, "landmask_imgs.pkl")
far = os.path.join(basepath, "atmosphericriver_TMQ.h5")
dataset = AR(fland=fland, far=far, repo_path=repo_path)
# metrics = {'train':[LogLossMean(), MisclassRate(), MSE()], 'validation':[LogLossMean(), MisclassRate(), MSE()], 'test':[]}
experiment = FitPredictErrorExperiment(model=model_gen(nofm_layer1, fshape_layer1, learning_rate_layer1, coef_layer1,
fshape_layer2, stride_layer2, nofm_layer3, fshape_layer3,
learning_rate_layer3, coef_layer3, fshape_layer4, stride_layer4,
learning_rate_layer5, coef_layer5, learning_rate_output, coef_output,
num_epochs, batch_size), backend=gen_backend(),dataset=dataset,
predictions = ['train', 'test'])
metrics = experiment.run()
result = float(metrics['test']['MisclassPercentage_TOP_1'] / 100)
# # Experiment result is dict: result[metric_set][metric_name]
# result = experiment.run()
# for setname in result.keys():
# for metricname in result[setname].keys():
# print '%s_%s: %f' % (setname, metricname, result[setname][metricname])
# loss = result['validation']['MSE']
# return loss
return {'hyperopt': result}
def run():
model = create_model(nin=784)
backend = gen_backend(rng_seed=0)
dataset = MNIST(repo_path='~/data/')
experiment = FitPredictErrorExperiment(model=model,
backend=backend,
dataset=dataset)
experiment.run()
def run():
model = create_model(nin=784)
backend = gen_backend(rng_seed=0)
dataset = MNIST(repo_path='~/data/')
experiment = FitPredictErrorExperiment(model=model,
backend=backend,
dataset=dataset)
experiment.run()
def hyperopt(lrate, lrule_coef, num_epochs, batch_size):
basepath = "/Users/DOE6903584/NERSC/mantissa-new/AR/data"
fland = os.path.join(basepath, "landmask_imgs.pkl")
far = os.path.join(basepath, "atmosphericriver_TMQ.h5")
dataset = AR(fland=fland, far=far)
experiment = FitPredictErrorExperiment(model=model_gen(lrate, lrule_coef, num_epochs, batch_size), backend=gen_backend(),dataset=dataset)
metrics = experiment.run()
result = float(metrics['test']['MisclassPercentage_TOP_1'])
return {'hyperopt': result}
{'learning_rate': 0.01,
'weight_decay': 0.001,
'momentum_params':
{'coef': 0.9, 'type': 'constant'}
},
'type': 'gradient_descent_momentum_weight_decay'},
weight_init=val_init.UniformValGen(low=-0.1,high=0.1),
activation = Logistic()
)
)
layers.append(CostLayer(
name = 'cost',
ref_layer = layers[0],
cost = CrossEntropy()
)
)
model = MLP(num_epochs=10, batch_size=100, layers=layers)
return model
basepath = "/Users/DOE6903584/NERSC/mantissa-new/AR/data"
fland = os.path.join(basepath, "landmask_imgs.pkl")
far = os.path.join(basepath, "atmosphericriver_TMQ.h5")
dataset = AR(fland=fland, far=far)
experiment = FitPredictErrorExperiment(model=model_gen(), backend=gen_backend(),dataset=dataset)
experiment_run = experiment.run()
