def inverse_stoc_hyper_priors_test5():
"""
4 hyperparam, 6 params, mix of gamma, delta hyper, mix of gauss, laplace prior
"""
hyperprior_types = [9, 7, 9, 7]
prior_types = [4, 5, 4]
hyperprior_params = [[1., 2.], [1., 1.], [1., 5.], [3., 1.]]
prior_hyperparams = [0., 1., 2.]
hyper_dependence_lengths = [1, 3, 1, 1]
dependence_lengths = [1, 3, 2]
param_hyperprior_types = [1, 0, 3, 2]
param_prior_types = [1, 2, 0]
n_stoc = 4
n_dims = 6
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([0.1, 0.5, 0.6, 0.7, 0.8, 0.9, 0.4, 0.2, 0.1, 0.5])
prior(p)
v = [1., 1.6986436, 1.6986436, 1.6986436, 3., 2.03787088]
print isp.laplace_prior()(p[-6], 1., v[0])
print isp.gaussian_prior()(p[-5], 2., v[1])
print isp.gaussian_prior()(p[-4], 2., v[2])
print isp.gaussian_prior()(p[-3], 2., v[3])
print isp.gaussian_prior()(p[-2], 0., v[4])
print isp.gaussian_prior()(p[-1], 0., v[5])
def inverse_stoc_hyper_priors_test4():
"""
3 hyperparam, 4 params, gamma, delta, gamma hyper, gauss, laplace prior
"""
hyperprior_types = [9, 7, 9]
prior_types = [4, 5]
hyperprior_params = [[1., 2.], [1., 1.], [1., 5.]]
prior_hyperparams = [0., 1.]
hyper_dependence_lengths = [1, 2, 1]
dependence_lengths = [1, 3]
param_hyperprior_types = [0, 1, 2]
param_prior_types = [0, 1]
n_stoc = 3
n_dims = 4
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([0.1, 0.5, 0.6, 0.7, 0.8, 0.9, 0.4])
prior(p)
v = [4.35688457, 1., 1., 2.33597589]
print isp.gaussian_prior()(p[-4], 0., v[0])
print isp.laplace_prior()(p[-3], 1., v[1])
print isp.laplace_prior()(p[-2], 1., v[2])
print isp.laplace_prior()(p[-1], 1., v[3])
def inverse_stoc_hyper_priors_test2():
"""
single hyperparam, multiple params, gamma hyper, gauss prior
"""
hyperprior_types = [9]
prior_types = [4]
hyperprior_params = [[1., 2.]]
prior_hyperparams = [0.]
hyper_dependence_lengths = [1]
dependence_lengths = [1]
param_hyperprior_types = [0]
param_prior_types = [0]
n_stoc = 1
n_dims = 2
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([0.1, 0.6, 0.7])
prior(p)
print isp.gaussian_prior()(p[1:], 0., 4.35688457)
def inverse_stoc_hyper_priors_test3():
"""
2 hyperparam, multiple params, gamma and delta hyper, gauss prior
"""
hyperprior_types = [9, 7]
prior_types = [4]
hyperprior_params = [[1., 2.], [1., 1.]]
prior_hyperparams = [0.]
hyper_dependence_lengths = [1, 1]
dependence_lengths = [1]
param_hyperprior_types = [0, 1]
param_prior_types = [0]
n_stoc = 2
n_dims = 2
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([0.1, 0.5, 0.6, 0.7])
prior(p)
print isp.gaussian_prior()(p[-2], 0., 4.35688457)
print isp.gaussian_prior()(p[-1], 0., 1.)
def main(run_string):
###### load training data
data = 'bh_50'
data_suffix = '_tr_1.csv'
data_dir = '../../data/uci/'
data_prefix = data_dir + data
x_tr, y_tr = input_tools.get_x_y_tr_data(data_prefix, data_suffix)
x_tr = np.genfromtxt('../../data/linear_input_data.txt', delimiter=',')
y_tr = x_tr
batch_size = x_tr.shape[0]
###### get weight information
weights_dir = '../../data/' #for forward test
a1_size = 0
num_inputs = tools.get_num_inputs(x_tr)
num_outputs = tools.get_num_outputs(y_tr)
layer_sizes = [1, num_inputs] * 2
m_trainable_arr = [True, True] * 2 + [False]
b_trainable_arr = [True, True] * 2 + [False]
num_weights = tools.calc_num_weights3(num_inputs, layer_sizes, num_outputs,
m_trainable_arr, b_trainable_arr)
###### check shapes of training data
x_tr, y_tr = tools.reshape_x_y_twod(x_tr, y_tr)
###### setup prior
hyper_type = "deterministic" # "stochastic" or "deterministic"
var_type = "deterministic" # "stochastic" or "deterministic"
weight_shapes = tools.get_weight_shapes3(num_inputs, layer_sizes,
num_outputs, m_trainable_arr,
b_trainable_arr)
dependence_lengths = tools.get_degen_dependence_lengths(weight_shapes,
independent=True)
if hyper_type == "deterministic" and var_type == "deterministic":
prior_types = [4]
prior_hyperparams = [[0., 1.]]
param_prior_types = [0]
prior = inverse_priors.inverse_prior(prior_types, prior_hyperparams,
dependence_lengths,
param_prior_types, num_weights)
n_stoc = 0
n_stoc_var = 0
elif hyper_type == "stochastic" and var_type == "deterministic":
granularity = 'single'
hyper_dependence_lengths = tools.get_hyper_dependence_lengths(
weight_shapes, granularity)
hyperprior_types = [9]
prior_types = [4]
hyperprior_params = [[1. / 2., 1. / (2. * 100)]]
prior_hyperparams = [0.]
param_hyperprior_types = [0]
param_prior_types = [0]
n_stoc = len(hyper_dependence_lengths)
prior = isp.inverse_stoc_hyper_prior(
hyperprior_types, prior_types, hyperprior_params,
prior_hyperparams, hyper_dependence_lengths, dependence_lengths,
param_hyperprior_types, param_prior_types, n_stoc, num_weights)
n_stoc_var = 0
elif hyper_type == "stochastic" and var_type == "stochastic":
granularity = 'single'
hyper_dependence_lengths = tools.get_hyper_dependence_lengths(
weight_shapes, granularity)
var_dependence_lengths = [1]
n_stoc_var = len(var_dependence_lengths)
hyperprior_types = [9]
var_prior_types = [10]
prior_types = [4]
hyperprior_params = [[1. / 2., 1. / (2. * 100)]]
var_prior_params = [[1. / 2., 1. / (2. * 100)]]
prior_hyperparams = [0.]
param_hyperprior_types = [0]
var_param_prior_types = [0]
param_prior_types = [0]
n_stoc = len(hyper_dependence_lengths)
prior = isvp.inverse_stoc_var_hyper_prior(
hyperprior_types, var_prior_types, prior_types, hyperprior_params,
var_prior_params, prior_hyperparams, hyper_dependence_lengths,
var_dependence_lengths, dependence_lengths, param_hyperprior_types,
var_param_prior_types, param_prior_types, n_stoc, n_stoc_var,
num_weights)
###### test prior output from nn setup
if "nn_prior_test" in run_string:
prior_tests.nn_prior_test(prior, n_stoc + n_stoc_var + num_weights)
#set up np model
np_nn = npms.mlp_ResNet_2
npm = np_model(np_nn, x_tr, y_tr, batch_size, layer_sizes, m_trainable_arr,
b_trainable_arr, n_stoc_var)
ll_type = 'gauss' # 'gauss', 'av_gauss', 'categorical_crossentropy', 'av_categorical_crossentropy'
npm.setup_LL(ll_type)
###### test llhood output
if "forward_test_linear" in run_string:
forward_tests.forward_test_linear([npm], num_weights + n_stoc_var,
weights_dir)
###### setup polychord
nDerived = 0
settings = PyPolyChord.settings.PolyChordSettings(
n_stoc + n_stoc_var + num_weights, nDerived)
settings.base_dir = './np_chains/'
settings.file_root = data + "_slp_sh_sv_sm"
settings.nlive = 1000
###### run polychord
if "polychord1" in run_string:
PyPolyChord.run_polychord(npm, n_stoc, n_stoc_var, num_weights,
nDerived, settings, prior,
polychord_tools.dumper)
def inverse_stoc_hyper_priors_test14():
"""
real nn arch with 16 nn params, two hyperparams, input_size granularity,
degen dependent params
"""
num_inputs = 1
layer_sizes = [3, 2]
num_outputs = 1
print "num weights"
n_dims = tools.calc_num_weights(num_inputs, layer_sizes, num_outputs)
print tools.calc_num_weights(num_inputs, layer_sizes, num_outputs)
weight_shapes = tools.get_weight_shapes(num_inputs, layer_sizes,
num_outputs)
print "weight shapes"
print tools.get_weight_shapes(num_inputs, layer_sizes, num_outputs)
granularity = 'input_size'
hyper_dependence_lengths = tools.get_hyper_dependence_lengths(
weight_shapes, granularity)
n_stoc = len(hyper_dependence_lengths)
print "n_stoc"
print n_stoc
print "granularity"
print tools.get_hyper_dependence_lengths(weight_shapes, granularity)
print "number of weights per layer"
print tools.calc_num_weights_layers(weight_shapes)
dependence_lengths = tools.get_degen_dependence_lengths(weight_shapes)
print "degen dependence lengths"
print tools.get_degen_dependence_lengths(weight_shapes)
hyperprior_types = [9, 7]
prior_types = [4, 5]
hyperprior_params = [[1., 2.], [2., 0.]]
prior_hyperparams = [0., 1.]
param_hyperprior_types = [0, 1, 0, 0, 1, 1, 1, 0, 0]
param_prior_types = [0, 1, 0, 1, 0, 0, 1, 1, 0]
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([
0.1, 0.5, 0.6, 0.7, 0.8, 0.9, 0.4, 0.2, 0.1, 0.5, 0.7, 0.8, 0.9, 0.4,
0.2, 0.1, 0.5, 0.9, 0.2, 0.7, 0.4, 0.9, 0.3, 0.5, 0.6, 0.8
])
prior(p)
u = [0., 0., 0., 1., 1., 1., 0., 0., 1., 1., 0., 0., 0., 0., 1., 1., 0.]
v = [
4.35688457, 4.35688457, 4.35688457, 2., 2., 2., 1.47740087, 1.47740087,
1.28886271, 1.28886271, 2., 2., 2., 2., 2., 2.9938003, 4.35688457
]
print isp.gaussian_prior()(p[9], u[0], v[0])
print isp.gaussian_prior()(p[10], u[1], v[1])
print isp.gaussian_prior()(p[11], u[2], v[2])
print isp.laplace_prior()(p[12], u[3], v[3])
print isp.laplace_prior()(p[13], u[4], v[4])
print isp.laplace_prior()(p[14], u[5], v[5])
print isp.gaussian_prior()(p[15], u[6], v[6])
print isp.gaussian_prior()(p[16], u[7], v[7])
print isp.laplace_prior()(p[17], u[8], v[8])
print isp.laplace_prior()(p[18], u[9], v[9])
print isp.gaussian_prior()(p[19], u[10], v[10])
print isp.gaussian_prior()(p[20], u[11], v[11])
print isp.gaussian_prior()(p[21], u[12], v[12])
print isp.gaussian_prior()(p[22], u[13], v[13])
print isp.laplace_prior()(p[23], u[14], v[14])
print isp.laplace_prior()(p[24], u[15], v[15])
print isp.gaussian_prior()(p[25], u[16], v[16])
def inverse_stoc_hyper_priors_test9():
"""
real nn arch with 16 nn params, two hyperparams, layer granularity,
independent params
"""
num_inputs = 1
layer_sizes = [3, 2]
num_outputs = 1
print "num weights"
n_dims = tools.calc_num_weights(num_inputs, layer_sizes, num_outputs)
print tools.calc_num_weights(num_inputs, layer_sizes, num_outputs)
weight_shapes = tools.get_weight_shapes(num_inputs, layer_sizes,
num_outputs)
print "weight shapes"
print tools.get_weight_shapes(num_inputs, layer_sizes, num_outputs)
granularity = 'layer'
hyper_dependence_lengths = tools.get_hyper_dependence_lengths(
weight_shapes, granularity)
n_stoc = len(hyper_dependence_lengths)
print "number of weights per layer"
print tools.calc_num_weights_layers(weight_shapes)
dependence_lengths = tools.get_degen_dependence_lengths(weight_shapes,
independent=True)
print "degen dependence lengths"
print tools.get_degen_dependence_lengths(weight_shapes, independent=True)
hyperprior_types = [9, 7]
prior_types = [4]
hyperprior_params = [[1., 2.], [2., 0.]]
prior_hyperparams = [0.]
param_hyperprior_types = [1, 0, 1]
param_prior_types = [0]
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([
0.1, 0.5, 0.6, 0.7, 0.8, 0.9, 0.4, 0.2, 0.1, 0.5, 0.7, 0.8, 0.9, 0.4,
0.2, 0.1, 0.5, 0.9, 0.2, 0.7
])
prior(p)
u = [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]
v = [
2., 2., 2., 2., 2., 2., 1.6986436, 1.6986436, 1.6986436, 1.6986436,
1.6986436, 1.6986436, 1.6986436, 1.6986436, 2., 2., 2.
]
print isp.gaussian_prior()(p[3], u[0], v[0])
print isp.gaussian_prior()(p[4], u[1], v[1])
print isp.gaussian_prior()(p[5], u[2], v[2])
print isp.gaussian_prior()(p[6], u[3], v[3])
print isp.gaussian_prior()(p[7], u[4], v[4])
print isp.gaussian_prior()(p[8], u[5], v[5])
print isp.gaussian_prior()(p[9], u[6], v[6])
print isp.gaussian_prior()(p[10], u[7], v[7])
print isp.gaussian_prior()(p[11], u[8], v[8])
print isp.gaussian_prior()(p[12], u[9], v[9])
print isp.gaussian_prior()(p[13], u[10], v[10])
print isp.gaussian_prior()(p[14], u[11], v[11])
print isp.gaussian_prior()(p[15], u[12], v[12])
print isp.gaussian_prior()(p[16], u[13], v[13])
print isp.gaussian_prior()(p[17], u[14], v[14])
print isp.gaussian_prior()(p[18], u[15], v[15])
print isp.gaussian_prior()(p[19], u[16], v[16])
def inverse_stoc_hyper_priors_test7():
"""
real nn arch with 16 nn params, one hyperparam, single granularity,
degen dependent params (2 priors)
"""
num_inputs = 1
layer_sizes = [3, 2]
num_outputs = 1
print "num weights"
n_dims = tools.calc_num_weights(num_inputs, layer_sizes, num_outputs)
print tools.calc_num_weights(num_inputs, layer_sizes, num_outputs)
weight_shapes = tools.get_weight_shapes(num_inputs, layer_sizes,
num_outputs)
print "weight shapes"
print tools.get_weight_shapes(num_inputs, layer_sizes, num_outputs)
granularity = 'single'
hyper_dependence_lengths = tools.get_hyper_dependence_lengths(
weight_shapes, granularity)
n_stoc = len(hyper_dependence_lengths)
print "number of weights per layer"
print tools.calc_num_weights_layers(weight_shapes)
dependence_lengths = tools.get_degen_dependence_lengths(weight_shapes)
print "degen dependence lengths"
print tools.get_degen_dependence_lengths(weight_shapes)
hyperprior_types = [9]
prior_types = [4, 5]
hyperprior_params = [[1., 2.]]
prior_hyperparams = [0., 1.]
param_hyperprior_types = [0]
param_prior_types = [0, 1, 0, 1, 0, 0, 1, 1, 0]
prior = isp.inverse_stoc_hyper_prior(hyperprior_types, prior_types,
hyperprior_params, prior_hyperparams,
hyper_dependence_lengths,
dependence_lengths,
param_hyperprior_types,
param_prior_types, n_stoc, n_dims)
p = np.array([
0.1, 0.5, 0.6, 0.7, 0.8, 0.9, 0.4, 0.2, 0.1, 0.5, 0.7, 0.8, 0.9, 0.4,
0.2, 0.1, 0.5, 0.9
])
prior(p)
u = [0., 0., 0., 1., 1., 1., 0., 0., 1., 1., 0., 0., 0., 0., 1., 1., 0.]
v = [
4.35688457, 4.35688457, 4.35688457, 4.35688457, 4.35688457, 4.35688457,
4.35688457, 4.35688457, 4.35688457, 4.35688457, 4.35688457, 4.35688457,
4.35688457, 4.35688457, 4.35688457, 4.35688457, 4.35688457
]
print isp.gaussian_prior()(p[1], u[0], v[0])
print isp.gaussian_prior()(p[2], u[1], v[1])
print isp.gaussian_prior()(p[3], u[2], v[2])
print isp.laplace_prior()(p[4], u[3], v[3])
print isp.laplace_prior()(p[5], u[4], v[4])
print isp.laplace_prior()(p[6], u[5], v[5])
print isp.gaussian_prior()(p[7], u[6], v[6])
print isp.gaussian_prior()(p[8], u[7], v[7])
print isp.laplace_prior()(p[9], u[8], v[8])
print isp.laplace_prior()(p[10], u[9], v[9])
print isp.gaussian_prior()(p[11], u[10], v[10])
print isp.gaussian_prior()(p[12], u[11], v[11])
print isp.gaussian_prior()(p[13], u[12], v[12])
print isp.gaussian_prior()(p[14], u[13], v[13])
print isp.laplace_prior()(p[15], u[14], v[14])
print isp.laplace_prior()(p[16], u[15], v[15])
print isp.gaussian_prior()(p[17], u[16], v[16])
def main(run_string):
###### load training data
data = 'simple_tanh'
data_dir = '../../data/'
data_prefix = data_dir + data
x_tr, y_tr = input_tools.get_x_y_tr_data(data_prefix)
batch_size = x_tr.shape[0]
###### get weight information
a1_size = 2
layer_sizes = [a1_size]
m_trainable_arr = [True, False]
b_trainable_arr = [False, False]
num_inputs = tools.get_num_inputs(x_tr)
num_outputs = tools.get_num_outputs(y_tr)
num_weights = tools.calc_num_weights3(num_inputs, layer_sizes, num_outputs,
m_trainable_arr, b_trainable_arr)
###### check shapes of training data
x_tr, y_tr = tools.reshape_x_y_twod(x_tr, y_tr)
###### setup keras model
model = kms.slp_model(num_inputs, num_outputs, layer_sizes)
km = kf.keras_model(model, x_tr, y_tr, batch_size)
loss = 'mse' # 'squared_error', 'av_squared_error', 'categorical_crossentropy', 'av_categorical_crossentropy'
km.setup_LL(loss)
#setup tf graph
tf_graph = tfgs.slp_graph
tfm = tff.tf_model(tf_graph, x_tr, y_tr, batch_size, layer_sizes,
m_trainable_arr, b_trainable_arr)
fit_metric = 'chisq' # 'chisq', 'av_chisq', 'categorical_crossentropy', 'av_categorical_crossentropy'
tfm.setup_LL(fit_metric)
#set up np model
np_nn = npms.slp_nn
npm = npf.np_model(np_nn, x_tr, y_tr, batch_size, layer_sizes,
m_trainable_arr, b_trainable_arr)
ll_type = 'gauss' # 'gauss', 'av_gauss', 'categorical_crossentropy', 'av_categorical_crossentropy'
npm.setup_LL(ll_type)
###### test llhood output
if "k_forward_test_linear" in run_string:
forward_tests.forward_test_linear([km], num_weights, data_dir)
if "tf_forward_test_linear" in run_string:
forward_tests.forward_test_linear([tfm], num_weights, data_dir)
if "np_forward_test_linear" in run_string:
forward_tests.forward_test_linear([npm], num_weights, data_dir)
###### setup prior
if hyper_type == "deterministic":
prior_types = [4]
prior_hyperparams = [[0., 1.]]
param_prior_types = [0]
prior = inverse_priors.inverse_prior(prior_types, prior_hyperparams,
dependence_lengths,
param_prior_types, num_weights)
n_stoc = 0
elif hyper_type == "stochastic":
granularity = 'single'
hyper_dependence_lengths = tools.get_hyper_dependence_lengths(
weight_shapes, granularity)
hyperprior_types = [9]
prior_types = [4]
hyperprior_params = [[0.1 / 2., 0.1 / (2. * 100)]]
prior_hyperparams = [0.]
param_hyperprior_types = [0]
param_prior_types = [0]
n_stoc = len(hyper_dependence_lengths)
prior = isp.inverse_stoc_hyper_prior(
hyperprior_types, prior_types, hyperprior_params,
prior_hyperparams, hyper_dependence_lengths, dependence_lengths,
param_hyperprior_types, param_prior_types, n_stoc, num_weights)
###### test prior output from nn setup
if "nn_prior_test" in run_string:
prior_tests.nn_prior_test(prior, n_stoc + num_weights)
###### setup polychord
nDerived = 0
settings = PyPolyChord.settings.PolyChordSettings(n_stoc + num_weights,
nDerived)
settings.file_root = data + "slp_1"
settings.nlive = 1000
###### run polychord
if "k_polychord1" in run_string:
settings.base_dir = './keras_chains/'
PyPolyChord.run_polychord(km, num_weights, nDerived, settings, prior,
polychord_tools.dumper)
if "tf_polychord1" in run_string:
settings.base_dir = './tf_chains/'
PyPolyChord.run_polychord(tfm, num_weights, nDerived, settings, prior,
polychord_tools.dumper)
if "np_polychord1" in run_string:
settings.base_dir = './np_chains/'
PyPolyChord.run_polychord(npm, num_weights, nDerived, settings, prior,
polychord_tools.dumper)
