SD_err = np.nanstd(np.squeeze(all_SD), axis=0) #/np.sqrt(len(all_SD))
output_dics = []
for d in task.positives:
output_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in pos_conds])[0][0])
input_dics = []
for d in input_task.positives:
input_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in pos_conds])[0][0])
dicplt.dichotomy_plot(PS,
CCGP,
SD,
input_dics=input_dics,
output_dics=output_dics,
other_dics=[pos_conds.index((0, 2, 5, 7))],
out_MI=out_MI.mean(0))
#%%
if two_layers:
z1 = nonlinearity(torch.matmul(W1, inputs.T) + b1).detach().numpy()
z = nonlinearity(torch.matmul(W2, torch.tensor(z1)) +
b2).detach().numpy().T
else:
z = nonlinearity(torch.matmul(W1, inputs.T) + b1).detach().numpy().T
x_ = np.stack([
inputs[inp_condition == i, :].mean(0).detach().numpy()
for i in np.unique(conds)
SD_err = np.nanstd(np.squeeze(all_SD), axis=0)#/np.sqrt(len(all_SD))
output_dics = []
for d in task.positives:
output_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in pos_conds])[0][0])
if 'bits' in dir(task):
input_dics = []
for d in task.bits:
input_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in pos_conds])[0][0])
else:
input_dics = None
dicplt.dichotomy_plot(PS, CCGP, SD, PS_err=PS_err, CCGP_err=CCGP_err, SD_err=SD_err,
input_dics=input_dics, output_dics=output_dics,
out_MI=out_MI.mean(0))
else:
nrow = int(np.sqrt(len(all_PS)))
ncol = len(all_PS)//nrow
plt.figure()
for i, (PS, CCGP, SD) in enumerate(zip(all_PS, almost_all_CCGP, all_SD)):
output_dics = []
for d in task.positives:
output_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in dic_pos[i]])[0][0])
if 'bits' in dir(task):
input_dics = []
for i, (PS, CCGP, SD) in enumerate(zip(all_PS, almost_all_CCGP, all_SD)):
output_dics = []
for d in output_task.positives:
output_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in pos_conds])[0][0])
input_dics = []
for d in input_task.positives:
input_dics.append(np.where([(list(p) == list(d)) or (list(np.setdiff1d(range(num_cond),p))==list(d))\
for p in pos_conds])[0][0])
plt.subplot(nrow, ncol, i + 1)
dicplt.dichotomy_plot(PS,
CCGP,
SD,
input_dics=input_dics,
output_dics=output_dics,
other_dics=[pos_conds.index((0, 2, 5, 7))],
out_MI=out_MI[i],
include_legend=(i == 0),
include_cbar=False,
s=10)
if np.mod(i, ncol) > 0:
plt.yticks([])
plt.ylabel('')
if (i + 1) // nrow < nrow:
plt.xlabel('')
plt.xticks([])
#%%
n_compute = 10000
lag = 3