hpvis.finished_runs_over_time(all_runs) # This one visualizes the spearman rank correlation coefficients of the losses # between different budgets. hpvis.correlation_across_budgets(result) # For model based optimizers, one might wonder how much the model actually helped. # The next plot compares the performance of configs picked by the model vs. random ones hpvis.performance_histogram_model_vs_random(all_runs, id2conf) plt.show() d1 = res.get_pandas_dataframe()[0] loss = res.get_pandas_dataframe()[1] d1['loss'] = loss if False: result = res # get all executed runs all_runs = result.get_all_runs() # get the 'dict' that translates config ids to the actual configurations id2conf = result.get_id2config_mapping() lcs = result.get_learning_curves() hpvis.interactive_HBS_plot(lcs, tool_tip_strings=hpvis.default_tool_tips( result, lcs))
import hpbandster.core.result as hpres import hpbandster.visualization as hpvis result = search._res # + # get all executed runs all_runs = result.get_all_runs() # get the 'dict' that translates config ids to the actual configurations id2conf = result.get_id2config_mapping() lcs = result.get_learning_curves() hpvis.interactive_HBS_plot(lcs) # - result.get_all_runs()[0].info['test_score_mean'], result.get_all_runs()[0].loss # + # Here is how you get he incumbent (best configuration) inc_id = result.get_incumbent_id() # let's grab the run on the highest budget inc_runs = result.get_runs_by_id(inc_id) inc_run = inc_runs[-1] # We have access to all information: the config, the loss observed during #optimization, and all the additional information inc_loss = inc_run.loss