def get_best_k_params_dist(n_iter,k_test,beta,Ncorr,
n_systems,n_concepts,emb_dim,noise,n_epicentres,linearsep,sd):
all_dicts = []
for i_iter in range(n_iter):
sd = np.random.randint(0,10e5)
A,B,X_A,X_B,B_shuff,B_idx_map = emb.get_data(n_systems,n_concepts,emb_dim,noise,n_epicentres,linearsep,sd,bplot=False)
dict_run = get_best_k_run_dist(k_test,beta,A,B_shuff,B_idx_map,Ncorr,i_iter)
all_dicts.append(dict_run)
#Get all_accs_per_k
all_accs_per_k = []
for k in k_test:
arr_k = np.array([d[k] for d in all_dicts]) #all values for each k
all_accs_per_k.append(arr_k)
#Get best average
avgs = np.array([np.mean(np.array(accs)) for accs in all_accs_per_k])
best_idx = np.argmax(avgs)
best_k_avg = k_test[best_idx] #rad value with highest average
best_avg = np.mean(np.array(all_accs_per_k[best_idx])) #highest average
clear_output(wait=True)
return best_k_avg, best_avg , all_accs_per_k
def get_best_Ncorr_krad_params_dist(n_iter,
n_systems,n_concepts,emb_dim,noise,n_epicentres,linearsep,sd,
Ncorr_test,k_test = None,rad_test = None):
all_dicts = []
p_test = k_test if k_test is not None else rad_test
pairs = list(itertools.product(Ncorr_test,p_test))
for i_iter in range(n_iter):
sd = np.random.randint(0,10e5)
A,B,X_A,X_B,B_shuff,B_idx_map = emb.get_data(n_systems,n_concepts,emb_dim,noise,n_epicentres,linearsep,sd,bplot=False)
dict_run = get_best_Ncorr_krad_run_angles(i_iter,A,B_shuff,B_idx_map,Ncorr_test,k_test = k_test,rad_test = rad_test)
all_dicts.append(dict_run)
#Get all_accs_per_Ncorr_krad
all_accs_per_Ncorr_krad = []
for Ncorr,p in pairs:
arr_Ncorr_krad = np.array([d[(Ncorr,p)] for d in all_dicts]) #all values for each k
all_accs_per_Ncorr_krad.append(arr_Ncorr_krad)
#Get best average
avgs = np.array([np.mean(np.array(accs)) for accs in all_accs_per_Ncorr_krad])
best_idx = np.argmax(avgs)
best_Ncorr_krad = pairs[best_idx] #rad value with highest average
best_avg = np.mean(np.array(all_accs_per_Ncorr_krad[best_idx])) #highest average
clear_output(wait=True)
return best_Ncorr_krad, best_avg , all_accs_per_Ncorr_krad
def get_best_krad_param_angles(n_iter,Ncorr,
n_systems,n_concepts,emb_dim,noise,n_epicentres,linearsep,sd,
k_test = None,rad_test = None):
#Test either k or rad
all_dicts = []
for i_iter in range(n_iter):
sd = np.random.randint(0,10e5)
A,B,X_A,X_B,B_shuff,B_idx_map = emb.get_data(n_systems,n_concepts,emb_dim,noise,n_epicentres,linearsep,sd,bplot=False)
dict_run = get_best_krad_run_angles(i_iter,Ncorr,A,B_shuff,B_idx_map,k_test = k_test,rad_test = rad_test)
all_dicts.append(dict_run)
#Get all_accs_per_rad
all_accs_per_p = []
p_test = k_test if k_test is not None else rad_test
for p in p_test:
arr_p = np.array([d[p] for d in all_dicts]) #all values for each k
all_accs_per_p.append(arr_p)
#Get best average
avgs = np.array([np.mean(np.array(accs)) for accs in all_accs_per_p])
best_idx = np.argmax(avgs)
best_p = p_test[best_idx] #rad value with highest average
best_avg = np.mean(np.array(all_accs_per_p[best_idx])) #highest average
clear_output(wait=True)
return best_p, best_avg , all_accs_per_p