pyMC API 推定

推定

pymc.ADVI

自動微分変分推論(ADVI - Automatic Difference Variational Inference)

class pymc.ADVI(*args, **kwargs)

pymc.ASVGD

Amortizedスタイン変分勾配下降(ASVGD - Amortized Stein Variational Gradient Descent)

class pymc.ASVGD(approx=None, estimator=<class 'pymc.variational.operators.KSD'>, kernel=<pymc.variational.test_functions.RBF object>, **kwargs)

pymc.SVGD

スタイン変分勾配下降

class pymc.SVGD(n_particles=100, jitter=1, model=None, start=None, random_seed=None, estimator=<class 'pymc.variational.operators.KSD'>, kernel=<pymc.variational.test_functions.RBF object>, **kwargs)

pymc.FullRankADVI

フルランク自動微分変分推論

class pymc.FullRankADVI(*args, **kwargs)

pymc.ImplicitGradient

暗黙の勾配(ImplicitGradient)

class pymc.ImplicitGradient(approx, estimator=<class 'pymc.variational.operators.KSD'>, kernel=<pymc.variational.test_functions.RBF object>, **kwargs)

pymc.Inference

変分推論の基底クラス

class pymc.Inference(op, approx, tf, **kwargs)

pymc.KLqp

カルバック・ライブラー・ダイバージェンス(KL情報量)推論

class pymc.KLqp(approx, beta=1.0)

pymc.fit

pymc.fit(n=10000, method='advi', model=None, random_seed=None, start=None, start_sigma=None, inf_kwargs=None, **kwargs)

pymc.Empirical

class pymc.Empirical(trace=None, size=None, **kwargs)

pymc.FullRank

シングルグループフルランク近似

class pymc.FullRank(*args, **kwargs)

pymc.MeanField

class pymc.MeanField(*args, **kwargs)

pymc.sample_approx

pymc.sample_approx(approx, draws=100, include_transformed=True)

pymc.Approximation

グループ近似のラッパー

class pymc.Approximation(groups, model=None)

pymc.Group

class pymc.Group(group=None, vfam=None, params=None, *args, **kwargs)

pymc.variational.operators.KL

class pymc.variational.operators.KL(approx, beta=1.0)

pymc.variational.operators.KSD

カーネル化スタイン下降を元にした演算

class pymc.variational.operators.KSD(approx, temperature=1)

pymc.Stein

class pymc.Stein(approx, kernel=<pymc.variational.test_functions.RBF object>, use_histogram=True, temperature=1)

pymc.adadelta

Adaデルタ　更新

pymc.adadelta(loss_or_grads=None, params=None, learning_rate=1.0, rho=0.95, epsilon=1e-06)

(使用例)

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = adadelta(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = adadelta(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.adagrad

Adagrad 更新

pymc.adagrad(loss_or_grads=None, params=None, learning_rate=1.0, epsilon=1e-06)

（使用例）

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = adagrad(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = adagrad(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.adagrad_window

pymc.adagrad_window(loss_or_grads=None, params=None, learning_rate=0.001, epsilon=0.1, n_win=10)

pymc.adam

Adam更新

pymc.adam(loss_or_grads=None, params=None, learning_rate=0.001, beta1=0.9, beta2=0.999, epsilon=1e-08)

(使用例)

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = adam(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = adam(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.adamax

Adamax 更新

pymc.adamax(loss_or_grads=None, params=None, learning_rate=0.002, beta1=0.9, beta2=0.999, epsilon=1e-08)

（使用例）

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = adamax(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = adamax(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.apply_momentum

更新モーメンタムを返す

pymc.apply_momentum(updates, params=None, momentum=0.9)

pymc.apply_netserv_momentum

Neterovモーメンタムを含む更新dictionaryを返す

pymc.apply_nesterov_momentum(updates, params=None, momentum=0.9)

pymc.momentum

確率的勾配下降のモーメンタム更新

pymc.momentum(loss_or_grads=None, params=None, learning_rate=0.001, momentum=0.9)

(使用例)

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = momentum(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = momentum(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.nesterov_momentum

確率的勾配下降のNeterovモーメンタム更新

pymc.nesterov_momentum(loss_or_grads=None, params=None, learning_rate=0.001, momentum=0.9)

(使用例)

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = nesterov_momentum(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = nesterov_momentum(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.norm_comstraint

最大重み基準の制限と勾配畳み込み

pymc.norm_constraint(tensor_var, max_norm, norm_axes=None, epsilon=1e-07)

(使用例)

>>> param = pytensor.shared(
    np.random.randn(100, 200).astype(pytensor.config.floatX))
>>> update = param + 100
>>> update = norm_constraint(update, 10)
>>> func = pytensor.function([], [], updates=[(param, update)])
>>> # Apply constrained update
>>> _ = func()
>>> from lasagne.utils import compute_norms
>>> norms = compute_norms(param.get_value())
>>> np.isclose(np.max(norms), 10)
True

pymc.rmsprop

RMSProp 更新

pymc.rmsprop(loss_or_grads=None, params=None, learning_rate=1.0, rho=0.9, epsilon=1e-06)

(使用例)

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = rmsprop(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = rmsprop(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.sgd

確率的勾配下降(SGD)の更新

pymc.sgd(loss_or_grads=None, params=None, learning_rate=0.001)

(使用例)

>>> a = pytensor.shared(1.)
>>> b = a*2
>>> updates = sgd(b, [a], learning_rate=.01)
>>> isinstance(updates, dict)
True
>>> optimizer = sgd(learning_rate=.01)
>>> callable(optimizer)
True
>>> updates = optimizer(b, [a])
>>> isinstance(updates, dict)
True

pymc.total_norm_constraint

テンソルリストの再計量

pymc.total_norm_constraint(tensor_vars, max_norm, epsilon=1e-07, return_norm=False)

(使用例)

from lasagne.layers import InputLayer, DenseLayer
import lasagne
from lasagne.updates import sgd, total_norm_constraint
x = pt.matrix()
y = pt.ivector()
l_in = InputLayer((5, 10))
l1 = DenseLayer(l_in, num_units=7, nonlinearity=pt.special.softmax)
output = lasagne.layers.get_output(l1, x)
cost = pt.mean(pt.nnet.categorical_crossentropy(output, y))
all_params = lasagne.layers.get_all_params(l1)
all_grads = pt.grad(cost, all_params)
scaled_grads = total_norm_constraint(all_grads, 5)
updates = sgd(scaled_grads, all_params, learning_rate=0.1)