import numpy as np
from thor_client import ThorClient
from franke import franke
# Create experiment.
tc = ThorClient()
name = "Franke Function"
# Create space.
dims = [
{"name": "x", "dim_type": "linear", "low": 0., "high": 1.},
{"name": "y", "dim_type": "linear", "low": 0., "high": 1.}
]
exp = tc.create_experiment(name, dims)
# Main optimization loop.
for i in range(30):
# Request new recommendation.
rec = exp.create_recommendation()
x = rec.config
# Evaluate new recommendation.
val = franke(np.array([x["x"], x["y"]]))
# Submit recommendation.
rec.submit_recommendation(val)
{
"name": "C",
"dim_type": "integer",
"low": 0,
"high": 24
},
{
"name": "alpha",
"dim_type": "integer",
"low": 0,
"high": 13
},
{
"name": "epsilon",
"dim_type": "integer",
"low": 0,
"high": 3
},
]
exp = tc.create_experiment(name, dims, overwrite=True)
# Main optimization loop.
for i in range(100):
# Request new recommendation.
rec = exp.create_recommendation()
x = rec.config
# Evaluate new recommendation.
val = -svm_on_grid(x["C"], x["alpha"], x["epsilon"])
# Submit recommendation.
rec.submit_recommendation(val)
batch_xs, batch_ys = mnist.train.next_batch(n_batch)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
# At the end of the learning sequence, return the accuracy on the test
# set of images.
return sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})
# Create experiment.
tc = ThorClient()
dims = [
{"name": "learning rate", "dim_type": "logarithmic", "low": 1e-8, "high": 1.},
{"name": "regularizer", "dim_type": "linear", "low": 0., "high": 1.},
{"name": "batch size", "dim_type": "integer", "low": 20, "high": 2000},
{"name": "epochs", "dim_type": "integer", "low": 5, "high": 20000}
]
exp = tc.create_experiment("MNIST Logistic Regression", dims, overwrite=True)
# Main optimization loop.
for i in range(100):
# Request new recommendation.
rec = exp.create_recommendation()
c = rec.config
# Evaluate new recommendation.
val = mnist_logistic(
c["learning rate"], c["regularizer"], c["batch size"], c["epochs"]
)
# Submit recommendation.
rec.submit_recommendation(val)