neuralyzer / pyabc Goto Github PK

Add description to docs and make it easier to handle API wise.

Stop if due to model stochasticity the acceptance rate drops a lot. Include in the ABCSMC.run method.

I'm trying to get the weighted average of the parameters for each population in an ABC run, like this:

for pop in history.get_all_populations()['t'].unique():
    if pop < 0: 
        continue
    df_params, weights = history.get_distribution(m=0, t=pop)   # <-- pop is of type 'numpy.int64'
    mean_params = np.average(df_params.values, weights=weights, axis=0)
    print('population: {}, params: {}'.format(pop, mean_params))

However, this gives an error: ZeroDivisionError: Weights sum to zero, can't be normalized.

This can be solved by casting the argument to get_distribution to int like this (note the int(pop) argument:

df_params, weights = history.get_distribution(m=0, t=int(pop)) # <-- pop is of type 'int'

Thus, it appears that get_distribution() does not accept numpy.int64 as a valid argument although it is a valid integer. Perhaps this casting can be done internally? Alternatively, a TypeError exception should be raised.