def predict(self, test_inputs, batch_size=None):
"""
Predict outputs given inputs.
Parameters
----------
test_inputs : ndarray
Points on which we wish to make predictions. Dimensions: num_test * input_dim.
batch_size : int
The size of the batches we make predictions on. If batch_size is None, predict on the
entire test set at once.
Returns
-------
ndarray
The predicted mean of the test inputs. Dimensions: num_test * output_dim.
ndarray
The predicted variance of the test inputs. Dimensions: num_test * output_dim.
"""
if batch_size is None:
num_batches = 1
else:
num_batches = util.ceil_divide(test_inputs.shape[0], batch_size)
test_inputs = np.array_split(test_inputs, num_batches)
pred_means = util.init_list(0.0, [num_batches])
pred_vars = util.init_list(0.0, [num_batches])
for i in xrange(num_batches):
pred_means[i], pred_vars[i] = self.session.run(
self.predictions, feed_dict={self.test_inputs: test_inputs[i]})
return np.concatenate(pred_means, axis=0), np.concatenate(pred_vars,
axis=0)
def _build_interim_vals(self, kernel_chol, inducing_inputs, train_inputs):
kern_prods = util.init_list(0.0, [self.num_latent])
kern_sums = util.init_list(0.0, [self.num_latent])
for i in xrange(self.num_latent):
ind_train_kern = self.kernels[i].kernel(inducing_inputs[i, :, :],
train_inputs)
# Compute A = Kxz.Kzz^(-1) = (Kzz^(-1).Kzx)^T.
kern_prods[i] = tf.transpose(
tf.cholesky_solve(kernel_chol[i, :, :], ind_train_kern))
# We only need the diagonal components.
kern_sums[i] = (self.kernels[i].diag_kernel(train_inputs) -
util.diag_mul(kern_prods[i], ind_train_kern))
kern_prods = tf.stack(kern_prods, 0)
kern_sums = tf.stack(kern_sums, 0)
return kern_prods, kern_sums
def _build_sample_info(self, kern_prods, kern_sums, means, covars):
sample_means = util.init_list(0.0, [self.num_latent])
sample_vars = util.init_list(0.0, [self.num_latent])
for i in xrange(self.num_latent):
if self.diag_post:
quad_form = util.diag_mul(kern_prods[i, :, :] * covars[i, :],
tf.transpose(kern_prods[i, :, :]))
else:
full_covar = tf.matmul(covars[i, :, :],
tf.transpose(covars[i, :, :]))
quad_form = util.diag_mul(
tf.matmul(kern_prods[i, :, :], full_covar),
tf.transpose(kern_prods[i, :, :]))
sample_means[i] = tf.matmul(kern_prods[i, :, :],
tf.expand_dims(means[i, :], 1))
sample_vars[i] = tf.expand_dims(kern_sums[i, :] + quad_form, 1)
sample_means = tf.concat(sample_means, 1)
sample_vars = tf.concat(sample_vars, 1)
return sample_means, sample_vars
def _build_entropy(self, weights, means, covars):
# First build half a square matrix of normals. This avoids re-computing symmetric normals.
log_normal_probs = util.init_list(
0.0, [self.num_components, self.num_components])
for i in xrange(self.num_components):
for j in xrange(i, self.num_components):
for k in xrange(self.num_latent):
if self.diag_post:
normal = util.DiagNormal(
means[i, k, :], covars[i, k, :] + covars[j, k, :])
else:
if i == j:
# Compute chol(2S) = sqrt(2)*chol(S).
covars_sum = tf.sqrt(2.0) * covars[i, k, :, :]
else:
# TODO(karl): Can we just stay in cholesky space somehow?
covars_sum = tf.cholesky(
util.mat_square(covars[i, k, :, :]) +
util.mat_square(covars[j, k, :, :]))
normal = util.CholNormal(means[i, k, :], covars_sum)
log_normal_probs[i][j] += normal.log_prob(means[j, k, :])
# Now compute the entropy.
entropy = 0.0
for i in xrange(self.num_components):
weighted_log_probs = util.init_list(0.0, [self.num_components])
for j in xrange(self.num_components):
if i <= j:
weighted_log_probs[j] = tf.log(
weights[j]) + log_normal_probs[i][j]
else:
weighted_log_probs[j] = tf.log(
weights[j]) + log_normal_probs[j][i]
entropy -= weights[i] * util.logsumexp(
tf.stack(weighted_log_probs))
return entropy
def _build_predict(self, weights, means, covars, inducing_inputs,
kernel_chol, test_inputs):
kern_prods, kern_sums = self._build_interim_vals(
kernel_chol, inducing_inputs, test_inputs)
pred_means = util.init_list(0.0, [self.num_components])
pred_vars = util.init_list(0.0, [self.num_components])
for i in xrange(self.num_components):
covar_input = covars[i, :, :] if self.diag_post else covars[
i, :, :, :]
sample_means, sample_vars = self._build_sample_info(
kern_prods, kern_sums, means[i, :, :], covar_input)
pred_means[i], pred_vars[i] = self.likelihood.predict(
sample_means, sample_vars)
pred_means = tf.stack(pred_means, 0)
pred_vars = tf.stack(pred_vars, 0)
# Compute the mean and variance of the gaussian mixture from their components.
weights = tf.expand_dims(tf.expand_dims(weights, 1), 1)
weighted_means = tf.reduce_sum(weights * pred_means, 0)
weighted_vars = (tf.reduce_sum(weights *
(pred_means**2 + pred_vars), 0) -
tf.reduce_sum(weights * pred_means, 0)**2)
return weighted_means, weighted_vars
def load_data(self):
loaded_data = load_json(self.get_path(), dict, "Storage")
self.swaps = init_list(loaded_data["trades"], SwapTrade) if "trades" in loaded_data else []
self.locks = init_list(loaded_data["locks"], dict) if "locks" in loaded_data else []
self.history = init_list(loaded_data["history"], SwapTransaction) if "history" in loaded_data else []
self.addresses = init_list(loaded_data["addresses"], dict) if "addresses"in loaded_data else [{"name": "default", "addresses": []}]