diff options
Diffstat (limited to 'simulation/active_blocks.py')
| -rw-r--r-- | simulation/active_blocks.py | 50 |
1 files changed, 20 insertions, 30 deletions
diff --git a/simulation/active_blocks.py b/simulation/active_blocks.py index 569cb6c..a1f6e76 100644 --- a/simulation/active_blocks.py +++ b/simulation/active_blocks.py @@ -1,11 +1,12 @@ -import main as mn +import utils import theano from theano import tensor as tsr import blocks -import blocks.algorithms, blocks.main_loop, blocks.extensions.monitoring +from blocks import algorithms, main_loop +import blocks.extensions as be +import blocks.extensions.monitoring as bm import picklable_itertools import numpy as np -from six.moves import range import fuel import fuel.datasets import collections @@ -19,28 +20,16 @@ class LearnedDataset(fuel.datasets.Dataset): """ provides_sources = ('x', 's') - def __init__(self, node_p, graph, source=mn.var_source, **kwargs): + def __init__(self, node_p, graph, **kwargs): super(LearnedDataset, self).__init__(**kwargs) self.node_p = node_p self.graph = graph - self.n_cascades = 1 # nbr of cascades of total size approx = request - self.source = lambda graph, t : source(graph, t, self.node_p) + self.source = lambda graph: utils.random_source(graph, self.node_p) def get_data(self, state=None, request=None): - floatX = 'int8' - x_obs = np.empty((request, len(self.graph)), dtype=floatX) - s_obs = np.empty((request, len(self.graph)), dtype=floatX) - i = 0 - while i < request: - x_tmp, s_tmp = mn.build_cascade_list( - mn.simulate_cascades(self.n_cascades, self.graph, self.source), - collapse=True - ) - x_obs[i:i + len(x_tmp)] = x_tmp[:request - i] - s_obs[i:i + len(x_tmp)] = s_tmp[:request - i] - i += len(x_tmp) - self.n_cascades += 1 # learn optimal nbr in loop - self.n_cascades = max(1, self.n_cascades - 2) + # floatX = 'int8' + x_obs, s_obs = utils.simulate_cascades(request, self.graph, self.source) + return (x_obs, s_obs) @@ -115,8 +104,9 @@ def create_fixed_data_stream(n_cascades, graph, batch_size, shuffle=True): -shuffle (bool): shuffle minibatches but not within minibatch, else sequential (non-shuffled) batches are used """ - cascades = mn.build_cascade_list(mn.simulate_cascades(n_cascades, graph), - collapse=True) + cascades = utils.build_cascade_list( + utils.simulate_cascades(n_cascades, graph), + collapse=True) x_obs, s_obs = cascades[0], cascades[1] data_set = fuel.datasets.base.IndexableDataset(collections.OrderedDict( [('x', x_obs), ('s', s_obs)] @@ -138,23 +128,23 @@ def create_learned_data_stream(graph, batch_size): if __name__ == "__main__": batch_size = 1000 - graph = mn.create_star(1000) + graph = utils.create_wheel(1000) print('GRAPH:\n', graph, '\n-------------\n') x, s, params, cost = create_mle_model(graph) rmse, g_shared = rmse_error(graph, params) - alg = blocks.algorithms.GradientDescent( - cost=-cost, parameters=[params], step_rule=blocks.algorithms.AdaDelta() + alg = algorithms.GradientDescent( + cost=-cost, parameters=[params], step_rule=blocks.algorithms.AdaDelta() ) data_stream = create_learned_data_stream(graph, batch_size) - loop = blocks.main_loop.MainLoop( + loop = main_loop.MainLoop( alg, data_stream, extensions=[ - blocks.extensions.FinishAfter(after_n_batches = 10**4), - blocks.extensions.monitoring.TrainingDataMonitoring([cost, params, - rmse, g_shared], after_batch=True), - blocks.extensions.Printing(every_n_batches = 10), + be.FinishAfter(after_n_batches=10**4), + bm.TrainingDataMonitoring([cost, params, + rmse, g_shared], after_batch=True), + be.Printing(every_n_batches=10), ActiveLearning(data_stream.dataset), ] ) |