model.vars[data_type]["p_pred"].stats()["mean"], columns=["mean"], index=model.input_data.index
)
pred_ui = pandas.DataFrame(
model.vars[data_type]["p_pred"].stats()["95% HPD interval"],
columns=["lower", "upper"],
index=model.input_data.index,
)
obs = pandas.DataFrame(model.vars[data_type]["p_obs"].value, columns=["value"], index=model.input_data.index)
# subset only test data
pred_test = pred.ix[test_ix]
obs_test = obs.ix[test_ix]
pred_ui_test = pred_ui.ix[test_ix]
# record statistics for test data
output["bias_" + rate_type] = mu.bias(pred_test, obs_test)
output["rmse_" + rate_type] = mu.rmse(pred_test, obs_test)
output["mae_" + rate_type] = mu.mae(pred_test, obs_test)
output["mare_" + rate_type] = mu.mare(pred_test, obs_test)
output["pc_" + rate_type] = mu.pc(pred_ui_test, obs_test)
output["time_" + rate_type] = elapsed
# save information
output.to_csv("/clustertmp/dismod/model_comparison_" + str(model_num) + rate_type + str(replicate) + ".csv")
# create and save conversion plots
dismod3.graphics.plot_acorr(model)
pl.savefig("/clustertmp/dismod/model_comparison_" + str(model_num) + rate_type + str(replicate) + "acorr.png")
dismod3.graphics.plot_trace(model)
pl.savefig("/clustertmp/dismod/model_comparison_" + str(model_num) + rate_type + str(replicate) + "trace.png")
def test_bias():
bias = mu.bias(pred, obs)
assert bias == 1