diff options
Diffstat (limited to 'simulation/vi_blocks.py')
| -rw-r--r-- | simulation/vi_blocks.py | 17 |
1 files changed, 8 insertions, 9 deletions
diff --git a/simulation/vi_blocks.py b/simulation/vi_blocks.py index b78375b..5deb6f6 100644 --- a/simulation/vi_blocks.py +++ b/simulation/vi_blocks.py @@ -51,10 +51,10 @@ def create_vi_model(n_nodes, n_samp=100): if __name__ == "__main__": - n_cascades = 10000 - batch_size = 10 + batch_size = 100 + frequency = 10 n_samples = 50 - graph = utils.create_random_graph(n_nodes=4) + graph = utils.create_random_graph(n_nodes=10) print('GRAPH:\n', graph, '\n-------------\n') x, s, mu, sig, cost = create_vi_model(len(graph), n_samples) @@ -65,17 +65,16 @@ if __name__ == "__main__": alg = algorithms.GradientDescent(cost=cost, parameters=[mu, sig], step_rule=step_rules) - data_stream = ub.fixed_data_stream(n_cascades, graph, batch_size, - shuffle=False) - # data_stream = ub.dynamic_data_stream(graph, batch_size) + data_stream = ub.dynamic_data_stream(graph, batch_size) loop = main_loop.MainLoop( alg, data_stream, log_backend="sqlite", extensions=[ be.FinishAfter(after_n_batches=10**4), - bm.TrainingDataMonitoring([cost, mu, sig, rmse], - every_n_batches=10), - be.Printing(every_n_batches=100, after_epoch=False), + bm.TrainingDataMonitoring([cost, rmse, mu], every_n_batches=frequency), + be.Printing(every_n_batches=frequency, after_epoch=False), + ub.JSONDump("logs/tmp.json", every_n_batches=10), + #ub.ActiveLearning(dataset=data_stream.dataset, params=graph) ] ) loop.run() |