def test_convolve_hrf():
# Create HRF to convolve with
dg = hrf.double_gamma(
width=32, TR=1, a1=6.0, a2=12., b1=0.9, b2=0.9, c=0.35
)
# -----------
# 1d convolve
# -----------
dm1 = np.zeros(32)
dm1[0] = 1
assert np.allclose(dm.convolve_hrf(dm1, dg), dg), (
"1d convolve doesn't match template")
assert np.allclose(dm.convolve_hrf(dm1, dg, [0]), dg), (
"1d convolve cols malfunction")
# -----------
# 2d convolve
# -----------
dm1 = np.zeros([32, 2])
dm1[0,:] = 1
# Basic 2d
assert np.allclose(dm.convolve_hrf(dm1, dg), np.vstack([dg,dg]).T), (
"2d doesn't match template")
# Parial dm
assert np.allclose(dm.convolve_hrf(dm1[:,0], dg), dg), (
"2d (col 0) convolve doesn't match template")
# Select cols
assert np.allclose(dm.convolve_hrf(dm1, dg, [0]), dg[:,np.newaxis]), (
"2d cols (0) malfunction")
assert np.allclose(dm.convolve_hrf(dm1, dg, [1]), dg[:,np.newaxis]), (
"2d cols (1) malfunction")
assert np.allclose(dm.convolve_hrf(dm1, dg, [0,1]), np.vstack([dg,dg]).T), (
"2d cols (0,1) malfunction")
# Create data
if args.behave == 'learn':
trials, acc, p, prng = behave.learn(
n_cond, args.n_trials,
loc=prng.normal(3, .3), prng=prng
)
elif args.behave == 'random':
trials, acc, p, prng = behave.random(
n_cond, args.n_trials, prng=prng
)
else:
raise ValueError('--behave not understood')
df, rlpars = reinforce.rescorla_wagner(trials, acc, p, prng=prng)
# Convolve with HRF
df = convolve_hrf(df, dg(), asbold)
# Orth select regressors
to_orth = [['box', bold] for bold in asbold if bold != 'box']
for orth in to_orth:
df[orth[1]+'_o'] = orthogonalize(df, orth)[orth[1]]
# Do the regressions
n_results = {}
for model_name, model, test, hypoth in zip(*model_configs):
for bold_name in asbold:
l = df.shape[0]
noi, prng = white(l, prng=prng)
df['bold'] = create_bold([df[bold_name].values], None, noi)
for bold_name in asbold:
l = df_bold.shape[0]
noi, prng = white(l, prng=prng)
df_bold['bold'] = create_bold(
[df_bold[bold_name].values], dg(), noi
)
# Create predictor
for alpha_pred in alphas:
df_pred, _ = reinforce.rescorla_wagner(
trials, acc, p,
alpha=alpha_pred, prng=prng
)
df_pred = convolve_hrf(df_pred, dg(), asbold)
# Orth select predictors
to_orth = [['box', too] for too in asbold if too != 'box']
for orth in to_orth:
df_pred[orth[1]+'_o'] = orthogonalize(
df_pred, orth
)[orth[1]]
# Hold bold constant in the predictor loops
df_pred['bold'] = df_bold['bold']
# Finally! Get modeling.
for model_name, model, test, hypoth in zip(*model_configs):
# Regress
for bold_name in asbold:
l = df_bold.shape[0]
noi, prng = white(l, prng=prng)
df_bold['bold'] = create_bold([df_bold[bold_name].values], dg(),
noi)
# Create predictor
for alpha_pred in alphas:
df_pred, _ = reinforce.rescorla_wagner(trials,
acc,
p,
alpha=alpha_pred,
prng=prng)
df_pred = convolve_hrf(df_pred, dg(), asbold)
# Orth select predictors
to_orth = [['box', too] for too in asbold if too != 'box']
for orth in to_orth:
df_pred[orth[1] + '_o'] = orthogonalize(df_pred,
orth)[orth[1]]
# Hold bold constant in the predictor loops
df_pred['bold'] = df_bold['bold']
# Finally! Get modeling.
for model_name, model, test, hypoth in zip(*model_configs):
# Regress
smo = smf.ols(model, data=df_pred).fit()
trial, acc, p, prng = behave.learn(
args.n_cond, args.n_trials,
loc=prng.normal(3, .3), prng=prng
)
elif args.behave == 'random':
trial, acc, p, prng = behave.random(
args.n_cond, args.n_trials, prng=prng
)
else:
raise ValueError('--behave not understood')
df, rlpars = reinforce.rescorla_wagner(
trial, acc, p, alpha=args.alpha, prng=prng
)
del df['rand']
l = trial.shape[0]
df['count'] = np.repeat(n, l)
df['index'] = np.arange(l, dtype=np.int)
dfs.append(df)
df = pd.concat(dfs, axis=0)
if args.convolve:
tocon = ['box', 'acc', 'p', 'rpe', 'value']
condf = convolve_hrf(df, dg(), tocon)
for con in tocon:
df[con] = condf[con]
df.to_csv(args.name, index=False, float_format='%.8f')
# Create data
if args.behave == 'learn':
trials, acc, p, prng = behave.learn(
n_cond, args.n_trials,
loc=prng.normal(3, .3), prng=prng
)
elif args.behave == 'random':
trials, acc, p, prng = behave.random(
n_cond, args.n_trials, prng=prng
)
else:
raise ValueError('--behave not understood')
df, rlpars = reinforce.rescorla_wagner(trials, acc, p, prng=prng)
# Convolve with HRF
df = convolve_hrf(df, dg(), asbold)
# Orth select regressors
to_orth = [['box', too] for too in asbold if too != 'box']
for orth in to_orth:
df[orth[1]+'_o'] = orthogonalize(df, orth)[orth[1]]
# Do the regressions
n_results = {}
for model_name, model, test, hypoth in zip(*model_configs):
for bold_name in asbold:
l = df.shape[0]
noi, prng = white(l, prng=prng)
# Make bold
df['bold'] = create_bold([df[bold_name].values], None, noi)
