Univariate

Import libraries

import numpy as np
import matplotlib.pyplot as plt
import synthia as syn

Load sample data

ds_true = syn.util.load_dataset(name='SAF-Synthetic')
# We only select one level as we are dealing with univariate data
ds_true = ds_true.sel(level=-1)
ds_true['temperature_fl'].plot.hist()
plt.xlabel('Air temperature in K');

Fit and generate using samples

generator = syn.CopulaDataGenerator()
# We fit and store the original samples and use them to recontuct new synthetic samples.
generator.fit(ds_true, copula=syn.IndependenceCopula())
N_SAMPLES = 1000
ds_synth_empirical = generator.generate(N_SAMPLES)
ds_synth_empirical['temperature_fl'].plot.hist()
plt.xlabel('Air temperature in K');

Fit and generate using parametrized distribution

generator = syn.CopulaDataGenerator()
# We now fit a parametric distribution instead of using the original samples (it may take some time)
dist_names = set(syn.DistributionParameterizer.get_dist_names())
# Remove all very slow distributions
dist_names -= set(['genexpon', 'levy_stable', 'recipinvgauss', 'vonmises', 'ncx2',
                   'kstwo', 'ncf', 'tukeylambda', 'nct', 'skewnorm', 'norminvgauss',
                   'gausshyper', 'geninvgauss', 'studentized_range'])
generator.fit(ds_true, copula=syn.IndependenceCopula(),
    parameterize_by=syn.DistributionParameterizer(dist_names, verbose=False))
ds_synth_param = generator.generate(1000)
ds_synth_param['temperature_fl'].plot.hist()
plt.xlabel('Air temperature in K');

Stretching and unifomization

ds_synth_stretched = generator.generate(1000, stretch_factor=2)
ds_synth_stretched['temperature_fl'].plot.hist()
plt.xlabel('Air temperature in K');

ds_synth_uniformization = generator.generate(1000, uniformization_ratio=0.5)
ds_synth_uniformization['temperature_fl'].plot.hist()
plt.xlabel('Air temperature in K');

Exporting and saving generated data

# As the original xarray structured is mainted we can convert or save the data in any of xarray's supoorted formats
ds_synth_param.to_dataframe()

	temperature_fl
column
0	302.163574
1	291.881622
2	293.075470
3	299.407440
4	263.578949
...	...
995	294.080078
996	290.117401
997	272.338257
998	272.479218
999	301.396637

1000 rows × 1 columns

# Or saving as a netCDF file
ds_synth_param.to_netcdf('ds_synth_param.nc')

Saving the generator

# Alternatively we can store the generator object into a pickle object, that we can later reload
import pickle
pickled = pickle.dumps(generator)
generator = pickle.loads(pickled)

# Or to a file
with open('generator.pkl', 'wb') as f:
    pickle.dump(generator, f)

with open('generator.pkl', 'rb') as f:
    generator = pickle.load(f)