Paper reproduction

Thread

supernnova.paper.superNNova_thread.SuperNNova_stats_and_plots(settings)[source]

Reproduce stats and plots used for SuperNNova paper. BEWARE: Selection is hardcoded

Parameters:

settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_thread.SuperNNova_stats_and_plots_thread(df, settings, plots=True, debug=False)[source]

Stats quoted in paper which are not in the latex tables and plots

Parameters:

  • df (pandas.DataFrame) – summary statistics df

  • settings (ExperimentSettings) – custom class to hold hyperparameters

  • plots (Boolean optional) – make pltos or only printout stats

  • debug (Boolean optional) – only print tasks

Returns:

stats as organized in paper figures (png) : figures for paper at settings.dump_dir/figures/ lightcurves (png): lightcurves used on paper at settings.dump_dir/lightcurves/modelname.*png

Return type:

printout

supernnova.paper.superNNova_thread.baseline(df, settings, plots, debug)[source]

Baseline RNN

supernnova.paper.superNNova_thread.bayesian(df, df_delta, df_delta_ood, settings, plots, debug)[source]

Bayesian RNNs: BBB and Variational

supernnova.paper.superNNova_thread.towards_cosmo(df, df_delta, df_delta_ood, settings, plots, debug)[source]

Towards cosmology

Metrics

supernnova.paper.superNNova_metrics.select_df(df, sel_criteria, data=None)[source]

Select a subsample of a pandas Dataframe Valid only for one to two selection criteria

Parameters:

  • df (Pandas.Dataframe) –

  • sel_criteria (list) – selection criteria

  • data (optional) – if source_data must be override

supernnova.paper.superNNova_metrics.get_delta_metrics(df_stats, settings)[source]

Difference between models in SuperNNova paper.

BEWARE: selection hard coded

Parameters:

  • df (pandas.DataFrame) – dataframe containing summary stats

  • settings (ExperimentSettings) – custom class to hold hyperparameters

Returns:

dataframe containing delta metrics

Return type:

df (pandas.DataFrame)

supernnova.paper.superNNova_metrics.create_accuracy_latex_tables(df, settings)[source]

Latex accuracy tables for paper.

BEWARE: Selection is hardcoded

Parameters:

  • df (pandas.DataFrame) – summary statistics df

  • settings (ExperimentSettings) – custom class to hold hyperparameters

Plots

supernnova.paper.superNNova_plots.create_sigma_df(df_grouped, class_=0)[source]

From grouped prediction df create a df with sigma values

supernnova.paper.superNNova_plots.plot_acc_vs_nsn(df, settings)[source]

Plot accuracy vs number of SNe used for training

Parameters:

  • df (DataFrame) – prediction dataframe

  • settings (ExperimentSettings) – custom class to hold hyperparameters

Returns:

figure (png)

supernnova.paper.superNNova_plots.plot_calibration(settings)[source]

Plot reliability diagram

Parameters:

settings (ExperimentSettings) – custom class to hold hyperparameters

Returns:

figure (png)

supernnova.paper.superNNova_plots.plot_confusion_matrix(settings, cm, classes, normalize=False, cmap=<matplotlib.colors.LinearSegmentedColormap object>, nameout=None)[source]

Plot confusion matrix Based on sklearn tutorial

Parameters:

  • settings (ExperimentSettings) – custom class to hold hyperparameters

  • cm (np.array) – confusion matrix

  • classes (list) – name of classes in cm

  • normalize (optional) – Normalization can be applied by setting normalize=True

  • cmap (colormap) –

  • nameout (optional) – out name for figure

Returns:

figure (png)

supernnova.paper.superNNova_plots.multiplot_violin_paper(df, fname, settings)[source]

Plot data properties as violin plots.

Far from optimized code: seaborn does not make this easy so added a lot of formatting using raw matplotlib commands

Parameters:

  • df (DataFrame) – prediction dataframe

  • fname (filename) –

  • settings (ExperimentSettings) – custom class to hold hyperparameters

Returns:

figure (png)

supernnova.paper.superNNova_plots.binned_2d(bin_centers, y_dic, xname, yname, label_list, color_sequence, MARKER_LIST, nameout, extra_line=None)[source]

Plot scatter plot

Parameters:

  • bin_centers (np.array) – x centers

  • y_dic (dict) – y values

  • xname (str) – xlabel

  • yname (str) – ylabel

  • label_list (list(str)) – for legend

  • color_sequence (list(str)) – color code

  • MARKER_LIST (list(str)) – marker code

  • nameout (str) – out name of figure (.png)

  • extra_line (optional) – extra line on 0 and 100 (y)

Returns:

figure (png)

supernnova.paper.superNNova_plots.plot_acc_matrix(acc_dic, n_meas_dic, settings, nameout)[source]

Plot accuracy matrix

Parameters:

  • acc_dic (dict) – with np.arrays of accuracy

  • n_meas_dic (dict) – with n measurements required for this accuracy

  • settings (ExperimentSettings) – custom class to hold hyperparameters

  • nameout (str) – outname for plot

Returns:

figure (png)

supernnova.paper.superNNova_plots.plot_HDres_histos_vs_z(df, nameout, threshold_var='class0', threshold_list=[0.5, 0.7, 0.9], threshold_sign='>')[source]

Plot Hubble diagram residuals and histograms selects class sample and performas HD

Parameters:

  • df (DataFrame) – predictions

  • nameout (str) – outname figure

  • threshold_var (str) – which class used for sample selection

  • threshold_list (list) – list of probability threshold for selecting sample

  • threshold_sign (str) – sign of the probability threshold (e.g. “>” )

Returns:

figure (png)

supernnova.paper.superNNova_plots.seaborn_formatting_mag(df, settings)[source]

Seaborn friendly formatting

Basic formatting and eliminating outliers (to avoid rejection by seaborn of pd.DataFrame)

Parameters:

  • df (DataFrame) – predictions

  • settings (ExperimentSettings) – custom class to hold hyperparameters

Returns:

reformatted

Return type:

df (DataFrame)

supernnova.paper.superNNova_plots.make_measurements_df(df, settings, group_bayesian=False)[source]

Obtain measurements :param df: predictions :type df: DataFrame :param settings: custom class to hold hyperparameters :type settings: ExperimentSettings :param group_bayesian: if BRNNs ar eused need to group predictions :type group_bayesian: Boolean

Returns:

with measurements and necessary keys to compute accuracy

Return type:

df (DataFrame)

supernnova.paper.superNNova_plots.distance_modulus(df)[source]

Add distance modulus :param df: with SALT2 fitted features :type df: DataFrame

Returns:

with distance modulus computed

Return type:

df (DataFrame)

supernnova.paper.superNNova_plots.sel_eff(merged, threshold, settings)[source]

Efficiency curve for different probabilities

Parameters:

df (DataFrame) – with SALT2 fitted features

Returns:

mean_bins (np.array) efficience (np.array)

supernnova.paper.superNNova_plots.datasets_plots(df, settings)[source]

Dataset violin plots peak magnitudes and redshift distributions of representative and non-representative datasets

Parameters:

  • df (DataFrame) – predictions

  • settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.performance_plots(settings)[source]
Performance: accuracy vs. size training set

Uses only Saltfit data and Baseline and Random Forest models

Parameters:

settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.purity_vs_z(df, model_name, settings)[source]

Purity and contamination redshift evolution

Parameters:

  • df (DataFrame) –

  • modelname (str) – name of model to be used

  • settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.cadence_acc_matrix(df, model_name, settings)[source]

Matrix with accuracy w.r. number of measurements in a band

Correlation between accuracy and a certain number of observations required per filter

Parameters:

  • df (DataFrame) –

  • modelname (str) – name of model to be used

  • settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.hubble_residuals(df, model_name, fits, settings)[source]

Hubble residuals for classified supernovae as type Ia

Uses SALT2 fits to compute the distance modulus, therefore list is not complete if fit failed.

Parameters:

  • df (DataFrame) –

  • modelname (str) – name of model to be used

  • fits (DataFrame) – information from SALT2 fit

  • settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.cnf_matrix(df, model_name, settings)[source]

Get confusion matrix from predictions

Parameters:

  • df (DataFrame) – predictions for a given model

  • modelname (str) – name of model to be used

  • settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.plot_speed_benchmark(dump_dir)[source]

Plot RNN inference speed benchmarks

N.B. You should have run the speed benchmarks with

  • Plot speed based on device

  • Plot speed based on choice of model (BBB, Variational or Baseline)

Parameters:

dump_dir (str) – Root folder where results are stored

supernnova.paper.superNNova_plots.create_OOD_classification_plots(df, list_models, settings)[source]

OOD classification for a model (ppercentages of predictions by class and uncertainties)

Parameters:

  • df (pandas.DataFrame) – summary statistics df

  • list – list with models to use

  • settings (ExperimentSettings) – custom class to hold hyperparameters

supernnova.paper.superNNova_plots.science_plots(settings, onlycnf=False)[source]

Plots for SuperNNova paper

Saved in settings.figures_dir. Need to provide prediction files and linked settings

Parameters:

settings (ExperimentSettings) – custom class to hold hyperparameters