def analyze_by_t2t(R, trl_ix, t2t):
perc = [0]+[np.percentile(t2t, i) for i in [25, 50, 75, 100]]
n_assemblies = R.shape[1]
f, ax = plt.subplots(nrows=n_assemblies)
labels = {}
for i, p in enumerate(perc[:-1]):
p2 = perc[i+1]
ix = np.nonzero(np.logical_and(t2t>p, t2t<=p2))[0]
labels[i] = ix
for a in range(n_assemblies):
tmp = []
for ii, x in enumerate(ix):
xx = np.nonzero(trl_ix == x)[0]
tmp.append(R[xx, a])
labels[i, a, 'trl'] = np.vstack((tmp))
m = np.mean(np.vstack((tmp)), axis=0)
sem = np.std(np.vstack((tmp)), axis=0)/len(ix)
ax[a].plot(m, color=cmap_list[i])
ax[a].fill_between(np.arange(len(m)), m-sem, m+sem, color=cmap_list[i], alpha=.5)
ax[a].plot(int(np.floor(p2*10)), m[int(np.floor(p2*10))-1], '.', markersize=20, color=cmap_list[i])
#Classify new trials:
test={}
train={}
from sklearn.lda import LDA
chance = {}
for a in range(n_assemblies):
lda = LDA()
lda.n_components = len(perc) - 1
X = []
Y = []
X_test = []
Y_test = []
ix_train = {}
ix_test = {}
for k in range(len(perc)-1):
n = len(labels[k])
ix = np.random.permutation(n)
ix_train[k] = ix[:n/2]
ix_test[k] = ix[n/2:]
X.append(labels[k, a, 'trl'][ix_train[k],:])
Y.append([k]*len(ix_train[k]))
X_test.append(labels[k, a, 'trl'][ix_test[k], :])
Y_test.append([k]*len(ix_test[k]))
Y_train = np.hstack((Y))
lda.fit(np.vstack((X)), Y_train)
y_true = lda.predict(np.vstack((X)))
train[a] = np.sum(y_true==np.hstack((Y)))/float(len(y_true))
y_pred = lda.predict(np.vstack(X_test))
test[a] = np.sum(y_pred == np.hstack((Y_test)))/float(len(y_pred))
chance[a] = []
for i in range(100):
ix = np.random.permutation(len(Y_train))
lda.fit(np.vstack((X)), Y_train[ix])
y_pred = lda.predict(np.vstack(X))
chance[a].append(np.sum(y_pred == Y_train[ix])/float(len(y_pred)))
plt.show()
return train, test, chance
