def test_lag_crp_block(data):
"""Test block-CRP analysis."""
crp = fr.lag_crp(data, lag_key='block', count_unique=True)
# recalls: [[2, 3, NaN], [3, 1, 3]]
# blocks: [[2, 2, NaN], [2, 1, 2]]
np.testing.assert_array_equal(crp['actual'], np.array([1, 1, 0]))
np.testing.assert_array_equal(crp['possible'], np.array([2, 1, 0]))
np.testing.assert_array_equal(crp['prob'], np.array([0.5, 1.0, np.nan]))
def test_lag_crp_query_output(data):
"""Test lag-CRP analysis with item output position filter."""
crp = fr.lag_crp(data, item_query='output > 1 or not recall')
np.testing.assert_array_equal(crp['actual'], np.zeros(5))
np.testing.assert_array_equal(crp['possible'], np.zeros(5))
expected = np.empty(5)
expected.fill(np.nan)
np.testing.assert_array_equal(crp['prob'], expected)
def test_lag_crp_cat(self):
crp = fr.lag_crp(self.data, test_key='task', test=lambda x, y: x == y)
actual = np.array([1, 0, 0, 0, 0])
possible = np.array([1, 0, 0, 0, 0])
prob = np.array([1., np.nan, np.nan, np.nan, np.nan])
np.testing.assert_array_equal(crp['actual'], actual)
np.testing.assert_array_equal(crp['possible'], possible)
np.testing.assert_array_equal(crp['prob'], prob)
def test_lag_crp_query(self):
crp = fr.lag_crp(self.data, item_query='input != 2')
actual = np.array([1, 0, 0, 0, 0])
possible = np.array([1, 0, 0, 0, 0])
prob = np.array([1., np.nan, np.nan, np.nan, np.nan])
np.testing.assert_array_equal(crp['actual'], actual)
np.testing.assert_array_equal(crp['possible'], possible)
np.testing.assert_array_equal(crp['prob'], prob)
def test_lag_crp(self):
crp = fr.lag_crp(self.data)
actual = np.array([1, 0, 0, 1, 0])
possible = np.array([1, 2, 0, 1, 0])
prob = np.array([1., 0., np.nan, 1., np.nan])
np.testing.assert_array_equal(crp['actual'], actual)
np.testing.assert_array_equal(crp['possible'], possible)
np.testing.assert_array_equal(crp['prob'], prob)
def test_lag_crp_cat(data):
"""Test lag-CRP conditional on within-category transitions."""
crp = fr.lag_crp(data, test_key='task', test=lambda x, y: x == y)
actual = np.array([1, 0, 0, 0, 0])
possible = np.array([1, 0, 0, 0, 0])
prob = np.array([1.0, np.nan, np.nan, np.nan, np.nan])
np.testing.assert_array_equal(crp['actual'], actual)
np.testing.assert_array_equal(crp['possible'], possible)
np.testing.assert_array_equal(crp['prob'], prob)
def test_lag_crp_query_input(data):
"""Test lag-CRP analysis with item input position filter."""
crp = fr.lag_crp(data, item_query='input != 2')
actual = np.array([1, 0, 0, 0, 0])
possible = np.array([1, 0, 0, 0, 0])
prob = np.array([1.0, np.nan, np.nan, np.nan, np.nan])
np.testing.assert_array_equal(crp['actual'], actual)
np.testing.assert_array_equal(crp['possible'], possible)
np.testing.assert_array_equal(crp['prob'], prob)
def test_lag_crp(data):
"""Test basic lag-CRP analysis."""
crp = fr.lag_crp(data)
actual = np.array([1, 0, 0, 1, 0])
possible = np.array([1, 2, 0, 1, 0])
prob = np.array([1.0, 0.0, np.nan, 1.0, np.nan])
np.testing.assert_array_equal(crp['actual'], actual)
np.testing.assert_array_equal(crp['possible'], possible)
np.testing.assert_array_equal(crp['prob'], prob)
def test_plot_lag_crp(data):
stat = fr.lag_crp(data)
g = fr.plot_lag_crp(stat)
plt.close()
assert isinstance(g, sns.FacetGrid)
def test_plot_lag_crp(data):
"""Test plotting a lag-CRP curve."""
stat = fr.lag_crp(data)
g = fr.plot_lag_crp(stat)
plt.close()
assert isinstance(g, sns.FacetGrid)
def visualize_fit(model_class,
parameters,
data,
data_query=None,
experiment_count=1000,
savefig=False):
"""
Apply organizational analyses to visually compare the behavior of the model with these parameters against
specified dataset.
"""
# generate simulation data from model
model = model_class(**parameters)
try:
model.experience(np.eye(model.item_count, model.item_count + 1, 1))
except ValueError:
model.experience(np.eye(model.item_count, model.item_count))
sim = []
for experiment in range(experiment_count):
sim += [[experiment, 0, 'study', i + 1, i]
for i in range(model.item_count)]
for experiment in range(experiment_count):
sim += [[experiment, 0, 'recall', i + 1, o]
for i, o in enumerate(model.free_recall())]
sim = pd.DataFrame(
sim, columns=['subject', 'list', 'trial_type', 'position', 'item'])
sim_data = fr.merge_free_recall(sim)
# generate simulation-based spc, pnr, lag_crp
sim_spc = fr.spc(sim_data).reset_index()
sim_pfr = fr.pnr(sim_data).query('output <= 1').reset_index()
sim_lag_crp = fr.lag_crp(sim_data).reset_index()
# generate data-based spc, pnr, lag_crp
data_spc = fr.spc(data).query(data_query).reset_index()
data_pfr = fr.pnr(data).query('output <= 1').query(
data_query).reset_index()
data_lag_crp = fr.lag_crp(data).query(data_query).reset_index()
# combine representations
data_spc['Source'] = 'Data'
sim_spc['Source'] = model_class.__name__
combined_spc = pd.concat([data_spc, sim_spc], axis=0)
data_pfr['Source'] = 'Data'
sim_pfr['Source'] = model_class.__name__
combined_pfr = pd.concat([data_pfr, sim_pfr], axis=0)
data_lag_crp['Source'] = 'Data'
sim_lag_crp['Source'] = model_class.__name__
combined_lag_crp = pd.concat([data_lag_crp, sim_lag_crp], axis=0)
# generate plots of result
# spc
g = sns.FacetGrid(dropna=False, data=combined_spc)
g.map_dataframe(sns.lineplot, x='input', y='recall', hue='Source')
g.set_xlabels('Serial position')
g.set_ylabels('Recall probability')
plt.title('P(Recall) by Serial Position Curve')
g.add_legend()
g.set(ylim=(0, 1))
if savefig:
plt.savefig('figures/{}_fit_spc.jpeg'.format(model_class.__name__),
bbox_inches='tight')
else:
plt.show()
#pdf
h = sns.FacetGrid(dropna=False, data=combined_pfr)
h.map_dataframe(sns.lineplot, x='input', y='prob', hue='Source')
h.set_xlabels('Serial position')
h.set_ylabels('Probability of First Recall')
plt.title('P(First Recall) by Serial Position')
h.add_legend()
h.set(ylim=(0, 1))
if savefig:
plt.savefig('figures/{}_fit_pfr.jpeg'.format(model_class.__name__),
bbox_inches='tight')
else:
plt.show()
# lag crp
max_lag = 5
filt_neg = f'{-max_lag} <= lag < 0'
filt_pos = f'0 < lag <= {max_lag}'
i = sns.FacetGrid(dropna=False, data=combined_lag_crp)
i.map_dataframe(lambda data, **kws: sns.lineplot(
data=data.query(filt_neg), x='lag', y='prob', hue='Source', **kws))
i.map_dataframe(lambda data, **kws: sns.lineplot(
data=data.query(filt_pos), x='lag', y='prob', hue='Source', **kws))
i.set_xlabels('Lag')
i.set_ylabels('Recall Probability')
plt.title('Recall Probability by Item Lag')
i.add_legend()
i.set(ylim=(0, 1))
if savefig:
plt.savefig('figures/{}_fit_crp.jpeg'.format(model_class.__name__),
bbox_inches='tight')
else:
plt.show()
def time_lag_crp(self):
crp = fr.lag_crp(self.data)