def test_fit():
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window.fit(model=models.model_hierarchical)
def test_estimate_posteriors_data_overlay_slope():
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window.fit(model=models.model_hierarchical)
chart = window.plot(x='neuron', independent_axes=False)
chart.display()
window.chart.facet(row='mouse').display()
def test_radon(
# add_data, add_box,
add_condition_slope,
add_group_slope,
# do_mean_over_trials,
do_make_change):
window = BayesRegression(df=df_radon,
y='radon',
treatment='floor',
condition=['county'])
# window.plot(x='county').facet(row='floor').display()
window.fit(
add_condition_slope=
add_condition_slope, # do_mean_over_trials=do_mean_over_trials,
add_group_slope=add_group_slope,
do_make_change=do_make_change,
n_draws=100,
num_chains=1,
num_warmup=100)
# window.plot().display()
window.plot(
x=':O', # add_data=add_data,
).display()
window.plot()
window.chart_posterior_kde.display()
# window.chart_data_boxplot.display() # Fold change won't work, bc uneven number of entries
assert len(window.charts) > 3 # Should include kde
def test_gpu():
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition=['neuron', 'mouse'],
group='mouse',
detail='i_trial')
window.fit(model=models.model_hierarchical,
do_make_change='subtract',
add_condition_slope=True,
add_group_slope=True,
use_gpu=True)
assert window.data_and_posterior.dropna(
subset=['mu_intercept_per_group center interval'
])['mouse'].unique().size == 4
def test_slopes(do_make_change, detail):
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse',
detail=detail)
# try:
window.fit(
model=models.model_hierarchical,
do_make_change=do_make_change,
)
# fold_change_index_cols=('stim', 'mouse', 'neuron_x_mouse'))
window.chart.display()
window.window.chart_data_box_detail.display()
def test_two_groups():
df = generate_spikes_stim_types(
mouse_response_slope=3,
n_trials=2,
n_neurons=3,
n_mice=4,
dur=2,
)
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition=['stim_strength', 'neuron_x_mouse'],
group='mouse',
group2='neuron')
window.fit(model=models.model_hierarchical, add_group2_slope=True)
def test_data_replacement1():
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron_x_mouse',
group='mouse',
detail='i_trial',
add_data=True)
window.fit(
model=models.model_hierarchical,
do_make_change='subtract',
add_condition_slope=True,
# add_group_slope=True
)
posterior_no_nan = window.data_and_posterior.dropna(
subset=['mu_intercept_per_group center interval'])
assert posterior_no_nan['mouse'].unique().size == 4
def test_plot_slopes():
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse',
add_data=True)
window.fit(model=models.model_hierarchical)
window.plot()
window.plot_BEST()
def test_facet():
# Slopes:
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window.fit(model=models.model_hierarchical)
window.plot(row='neuron', width=40)
window.plot(x='neuron').facet(column='mouse')
def test_inheritance():
window = BayesWindow(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window = BayesRegression(window)
window.fit(model=models.model_hierarchical)
chart = window.plot(x='neuron', independent_axes=False)
chart.display()
def test_chirp_data1(plot_from_data_and_posterior):
dfdata = pd.read_csv(Path('tests') / 'test_data' / 'chirp_power.csv')
window = BayesRegression(df=dfdata,
y='Log power',
treatment='stim_on',
condition=['Stim phase', 'Inversion'],
group='Subject')
window.fit(model=models.model_hierarchical,
num_chains=1,
n_draws=100,
num_warmup=100)
chart = window.plot(
x='Stim phase', # color='Fid',
add_data=plot_from_data_and_posterior,
independent_axes=True)
chart.display()
chart_data = pd.DataFrame.from_records(
list(json.loads(chart.to_json())['datasets'].values())[0])
assert (chart_data['Stim phase'].unique().astype(float) ==
dfdata['Stim phase'].unique()).all()
def test_estimate_posteriors_two_conditions():
df = generate_spikes_stim_types(
mouse_response_slope=3,
n_trials=2,
n_neurons=3,
n_mice=4,
dur=2,
)
regression = BayesRegression(
BayesWindow(df=df,
y='isi',
treatment='stim',
condition=['neuron', 'stim_strength'],
group='mouse',
add_data=True))
regression.fit(model=models.model_hierarchical,
fold_change_index_cols=None,
do_mean_over_trials=False)
for condition_name in regression.window.condition:
assert condition_name in regression.data_and_posterior.columns, f'{condition_name} not in window.condition'
chart = regression.plot(x='neuron', column='neuron', row='mouse')
chart.display()
def test_plot_slopes_2levelslope():
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron_x_mouse',
group='mouse',
add_data=True)
window.fit(model=models.model_hierarchical, add_group_slope=True)
window.plot().display()
def test_single_condition_nodata():
window = BayesRegression(df=dfl,
y='Log power',
treatment='stim',
group='mouse')
window.fit(model=models.model_hierarchical,
do_make_change='divide',
dist_y='normal')
alt.layer(*plot_posterior(
df=window.data_and_posterior,
y_title='Log power',
)).display()
window.plot(independent_axes=True).display()
def test_estimate_posteriors_data_overlay_indep_axes_slope(
add_data, add_data_plot, add_group_slope):
window = BayesRegression(df=df,
add_data=add_data,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window.fit(model=models.model_hierarchical,
add_group_slope=add_group_slope)
chart = window.plot(independent_axes=True, add_data=add_data_plot)
chart.display()
if add_group_slope and add_data_plot:
chart = window.facet(column='neuron', row='mouse')
else:
chart = window.facet(column='neuron')
chart.display()
def test_estimate_posteriors_slope(add_condition_slope):
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition=['neuron', 'mouse'],
group='mouse',
add_data=True)
window.fit(models.model_hierarchical,
add_condition_slope=add_condition_slope)
chart = window.plot(x='neuron', column='neuron', row='mouse')
chart.display()
chart = window.plot(x='neuron', column='neuron', row='mouse')
chart.display()
def test_chirp_data(force_correct_fold_change_index_cols):
dfdata = pd.read_csv(Path('tests') / 'test_data' / 'chirp_power.csv')
window = BayesRegression(
df=dfdata,
y='Log power',
treatment='stim_on',
condition=['Stim phase', 'Inversion', 'Brain region'],
group='Subject')
if force_correct_fold_change_index_cols:
window.fit(model=models.model_hierarchical,
fold_change_index_cols=[
'Brain region', 'Stim phase', 'stim_on', 'Fid',
'Subject', 'Inversion'
],
num_chains=1,
n_draws=100,
num_warmup=100)
else:
window.fit(model=models.model_hierarchical,
num_chains=1,
n_draws=100,
num_warmup=100)
window.plot(x='Stim phase', color='Fid', independent_axes=True).display()
def test_plot_slopes_intercepts(do_make_change):
window = BayesRegression(df=dfl,
y='Power',
treatment='stim',
group='mouse',
add_data=True)
# Fit:
window.fit(model=models.model_hierarchical,
add_group_intercept=True,
zscore_y=False,
add_group_slope=False,
robust_slopes=False,
do_make_change=do_make_change,
dist_y='gamma',
num_chains=1,
n_draws=100,
num_warmup=100)
window.plot_intercepts(x='mouse').display()
chart_intercepts = window.chart_posterior_intercept
chart_intercepts.display()
def test_estimate_posteriors_slope_uneven_n_data_per_condition():
# Trying to reproduce Uneven number of entries in conditions! Try setting do_take_mean=True
df, df_monster, index_cols, firing_rates = generate_fake_spikes(
n_trials=10,
n_neurons=3,
n_mice=4,
dur=2,
)
df = df.drop(df[(df['i_trial'] == 0) &
# (df['neuron_x_mouse'] == '0m0bayes') &
(df['stim'] == 0)].index)
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse',
add_data=True)
window.fit(models.model_hierarchical, do_make_change='divide')
chart = window.plot(x='neuron', column='neuron', row='mouse')
chart.display()
chart = window.plot(x='neuron', column='neuron', row='mouse')
chart.display()
def test_chirp_data2():
dfdata = pd.read_csv(Path('tests') / 'test_data' / 'chirp_power.csv')
window = BayesRegression(
df=dfdata,
y='Log power',
treatment='stim_on',
# condition=['Condition code'],
group='Subject',
add_data=True)
window.fit(
model=models.model_hierarchical,
fold_change_index_cols=[ # 'Condition code',
'Brain region', 'Stim phase', 'stim_on', 'Fid', 'Subject',
'Inversion'
],
do_mean_over_trials=True,
num_chains=1,
n_draws=100,
num_warmup=100)
window.plot(x='Stim phase',
color='Fid',
add_data=True,
independent_axes=True).display()
def facet(self, **kwargs):
from bayes_window import BayesRegression
return BayesRegression.facet(self, **kwargs)
assert window.data_and_posterior.dropna(
subset=['mu_intercept_per_group center interval'
])['mouse'].unique().size == 4
# def test_stim_strength():
# df = []
# for slope in np.linspace(4, 400, 4):
# df1 = generate_fake_lfp(mouse_response_slope=slope)[0]
# df1['stim_strength'] = slope
# df.append(df1)
# df = pd.concat(df)
# BayesRegression(df=df, 'Power', treatment='stim_strength', group='mouse', detail='i_trial').data_box_detail()
# window = BayesRegression(df=df, 'Power', treatment='stim_strength', condition='mouse', detail='i_trial')
# window.fit(add_condition_slope=True, center_intercept=True, dist_y='normal', num_chains=1, n_draws=100, num_warmup=100)
# window.chart
window = BayesRegression(df=df, y='isi', treatment='stim', group='mouse')
window.fit(model=models.model_hierarchical)
def test_extra_plots():
window.plot_intercepts().display()
window.plot_detail_minus_intercepts().display()
# def test_explore_models():
# window.explore_models(parallel=False)
# window.explore_model_kinds(parallel=False)
def fit(self,
do_make_change='divide',
add_interaction=False,
add_data=False,
formula=None,
add_group_intercept=True,
add_group_slope=False,
add_nested_group=False,
**kwargs):
# model = MixedLM(endog=self.window.data[self.window.y],
# exog=self.window.data[self.window.condition],
# groups=self.window.data[self.window.group],
# # exog_re=exog.iloc[:, 0]
# )
self.b_name = 'lme'
# dehumanize all columns and variable names for statsmodels:
[
self.window.data.rename({col: col.replace(" ", "_")},
axis=1,
inplace=True)
for col in self.window.data.columns
]
self.window.y = self.window.y.replace(" ", "_")
self.window.group = self.window.group.replace(" ", "_")
self.window.treatment = self.window.treatment.replace(" ", "_")
self.window.do_make_change = do_make_change
include_condition = False # in all but the following cases:
if self.window.condition[0]:
self.window.condition[0] = self.window.condition[0].replace(
" ", "_")
if len(self.window.condition) > 1:
self.window.condition[1] = self.window.condition[1].replace(
" ", "_")
if len(self.window.data[self.window.condition[0]].unique()) > 1:
include_condition = True
# condition = None # Preallocate
# Make formula
if include_condition and not formula:
if len(self.window.condition) > 1:
raise NotImplementedError(
f'conditions {self.window.condition}. Use combined_condition'
)
# This would need a combined condition dummy variable and an index of condition in patsy:
# formula = f"{self.window.y} ~ 1+ {self.window.condition}(condition_index) | {self.window.treatment}"
# Combined condition
# self.window.data, self._key = utils.combined_condition(self.window.data.copy(), self.window.condition)
# self.window.condition = ['combined_condition']
# self.window.original_data = self.window.data.copy()
# Make dummy variables for each level in condition:
self.window.data = pd.concat(
(self.window.data,
pd.get_dummies(self.window.data[self.window.condition[0]],
prefix=self.window.condition[0],
prefix_sep='__',
drop_first=False)),
axis=1)
dummy_conditions = [
cond for cond in self.window.data.columns
if cond[:len(self.window.condition[0]) +
2] == f'{self.window.condition[0]}__'
]
if add_group_intercept and not add_group_slope and not add_nested_group:
formula = f"{self.window.y} ~ (1|{self.window.group}) + {self.window.treatment}| {dummy_conditions[0]}"
# eg 'firing_rate ~ stim|neuron_x_mouse__0 +stim|neuron_x_mouse__1 ... + ( 1 |mouse )'
for dummy_condition in dummy_conditions[1:]:
formula += f" + {self.window.treatment}|{dummy_condition}"
elif add_group_intercept and add_group_slope and not add_nested_group:
formula = (
f"{self.window.y} ~ ({self.window.treatment}|{self.window.group}) "
f" + {self.window.treatment}| {dummy_conditions[0]}")
for dummy_condition in dummy_conditions[1:]:
formula += f" + {self.window.treatment}|{dummy_condition}"
elif add_group_intercept and add_group_slope and add_nested_group:
formula = (
f"{self.window.y} ~ ({self.window.treatment}|{self.window.group}) + "
f"{self.window.treatment}| {dummy_conditions[0]}:{self.window.group}"
)
for dummy_condition in dummy_conditions[1:]:
formula += f" + {self.window.treatment}|{dummy_condition}:{self.window.group}"
# if add_interaction:
# formula += f"+ {condition} * {self.window.treatment}"
elif self.window.group and not formula:
# Random intercepts and slopes (and their correlation): (Variable | Group)
formula = f'{self.window.y} ~ C({self.window.treatment}, Treatment) + (1 | {self.window.group})' # (1 | {self.window.group}) +
# Random intercepts and slopes (without their correlation): (1 | Group) + (0 + Variable | Group)
# formula += f' + (1 | {self.window.group}) + (0 + {self.window.treatment} | {self.window.group})'
elif not formula:
formula = f"{self.window.y} ~ C({self.window.treatment}, Treatment)"
print(f'Using formula {formula}')
result = sm.mixedlm(formula,
self.window.data,
groups=self.window.data[self.window.group]).fit()
print(result.summary().tables[1])
self.data_and_posterior = utils.scrub_lme_result(
result, include_condition, self.window.condition[0],
self.window.data, self.window.treatment)
if add_data:
raise NotImplementedError(f'No adding data to LME')
self.data_and_posterior = utils.add_data_to_lme(
do_make_change, include_condition, self.posterior,
self.window.condition[0], self.window.data, self.window.y,
self.levels, self.window.treatment)
# self.trace.posterior = utils.rename_posterior(self.trace.posterior, self.b_name,
# posterior_index_name='combined_condition',
# group_name=self.window.group, group2_name=self.window.group2)
#
# # HDI and MAP:
# self.posterior = {var: utils.get_hdi_map(self.trace.posterior[var],
# prefix=f'{var} '
# if (var != self.b_name) and (var != 'slope_per_condition') else '')
# for var in self.trace.posterior.data_vars}
#
# # Fill posterior into data
# self.data_and_posterior = utils.insert_posterior_into_data(posteriors=self.posterior,
# data=self.window.data.copy(),
# group=self.window.group)
#
#
# self.posterior = utils.recode_posterior(self.posterior, self.levels, self.window.data, self.window.original_data,
# self.window.condition)
from bayes_window.slopes import BayesRegression
self.charts = BayesRegression.plot(self)
return self
def random_tests():
# TODO make a notebook for this
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window.fit(model=models.model_hierarchical, num_chains=1)
window.plot(x='neuron',
color='mouse',
independent_axes=True,
finalize=True)
window.plot(independent_axes=False, x='neuron:O', color='mouse')
window.plot(add_box=False,
independent_axes=True,
x='neuron:O',
color='mouse')
window.plot(independent_axes=False, x='neuron:O', color='mouse')
chart = window.plot(independent_axes=True, x='neuron:O', color='mouse')
chart.display()
chart.resolve_scale(y='independent')
window.facet(column='neuron')
window = BayesRegression(df=df,
y='isi',
treatment='stim',
condition='neuron',
group='mouse')
window.fit(model=models.model_hierarchical, num_chains=1)
window.plot(x='neuron', color='i_trial')
window.plot() # x='Stim phase', color='Fid')#,independent_axes=True)
window.facet(column='neuron', row='mouse')
