def fit_psychfunc(df):
choicedat = df.groupby('signed_contrast').agg({
'choice': 'count',
'choice2': 'mean'
}).reset_index()
pars, L = psy.mle_fit_psycho(
choicedat.values.transpose(),
P_model='erf_psycho_2gammas',
parstart=np.array(
[choicedat['signed_contrast'].mean(), 20., 0.05, 0.05]),
parmin=np.array([choicedat['signed_contrast'].min(), 0., 0., 0.]),
parmax=np.array([choicedat['signed_contrast'].max(), 100., 1, 1]))
df2 = {
'bias': pars[0],
'threshold': pars[1],
'lapselow': pars[2],
'lapsehigh': pars[3]
}
df2 = pd.DataFrame(df2, index=[0])
# # add some stuff
# df2['easy_correct'] = df.loc[np.abs(
# df['signed_contrast'] > 50), 'correct'].mean(skipna=True)
# df2['zero_contrast'] = df.loc[np.abs(
# df['signed_contrast'] == 0), 'choice2'].mean(skipna=True)
# df2['median_rt'] = df['rt'].median(skipna=True)
# df2['mean_rt'] = df['rt'].mean(skipna=True)
# # number of trials per day
# df4 = df.groupby(['session_start_time'])['correct'].count().reset_index()
# df2['ntrials_perday'] = [df4['correct'].values]
df2['ntrials'] = df['choice'].count()
return df2
def ibl_psychometric(psy_df, ax=None, **kwargs):
"""Calculates and plot psychometic curve from dataframe
datajoint independent
assumes that there is not signed contrast in dataframe
INPUTS: Dataframe where index is the trial number
OUTPUTS: psychometic fit using IBL function from Miles and fit parameters"""
#1st calculate some useful data...
psy_df.loc[:, 'contrastRight'] = psy_df['contrastRight'].fillna(0)
psy_df.loc[:, 'contrastLeft'] = psy_df['contrastLeft'].fillna(0)
psy_df.loc[:, 'signed_contrasts'] = (psy_df['contrastRight'] -
psy_df['contrastLeft']) * 100
unique_signed_contrasts = sorted(psy_df['signed_contrasts'].unique())
right_choices = psy_df['choice'] == -1
psy_df.loc[:, 'right_choices'] = right_choices
total_trials = []
right_trials = []
for cont in unique_signed_contrasts:
matching = (psy_df['signed_contrasts'] == cont)
total_trials.append(np.sum(matching))
right_trials.append(np.sum(right_choices[matching]))
prop_right_trials = np.divide(right_trials, total_trials)
pars, L = psy.mle_fit_psycho(
np.vstack([unique_signed_contrasts, total_trials, prop_right_trials]),
P_model='erf_psycho_2gammas',
parstart=np.array([np.mean(unique_signed_contrasts), 20., 0.05, 0.05]),
parmin=np.array([np.min(unique_signed_contrasts), 0., 0., 0.]),
parmax=np.array([np.max(unique_signed_contrasts), 100., 1, 1]))
return pars, L
def fit_psychfunc(df):
choicedat = df.groupby('signedContrast').agg({'trial':'count', 'choice2':'mean'}).reset_index()
pars, L = psy.mle_fit_psycho(choicedat.values.transpose(), P_model='erf_psycho_2gammas',
parstart=np.array([choicedat['signedContrast'].mean(), 20., 0.05, 0.05]),
parmin=np.array([choicedat['signedContrast'].min(), 0., 0., 0.]),
parmax=np.array([choicedat['signedContrast'].max(), 100., 1, 1]))
df2 = {'bias':pars[0],'threshold':pars[1], 'lapselow':pars[2], 'lapsehigh':pars[3]}
return pd.DataFrame(df2, index=[0])
def fit_psychfunc(stim_levels, n_trials, proportion):
# Fit a psychometric function with two lapse rates
#
# Returns vector pars with [bias, threshold, lapselow, lapsehigh]
from ibl_pipeline.utils import psychofit as psy
assert (stim_levels.shape == n_trials.shape == proportion.shape)
if stim_levels.max() <= 1:
stim_levels = stim_levels * 100
pars, _ = psy.mle_fit_psycho(np.vstack(
(stim_levels, n_trials, proportion)),
P_model='erf_psycho_2gammas',
parstart=np.array([0, 20, 0.05, 0.05]),
parmin=np.array([-100, 5, 0, 0]),
parmax=np.array([100, 100, 1, 1]))
return pars
def plot_psychometric(df, color='black', ax=None, **kwargs):
"""
Plots psychometric data for a given DataFrame of behavioural trials
If the data contains more than six different contrasts (or > three per side)
the data are fit with an erf function. The x-axis is percent contrast and
the y-axis is the proportion of 'rightward choices', i.e. trials where the
subject turned the wheel clockwise to threshold.
Example:
df = alf.load_behaviour('2018-09-11_1_Mouse1', r'\\server\SubjectData')
plot_psychometric(df)
Args:
df (DataFrame): DataFrame constructed from an ALF trials object.
ax (Axes): Axes to plot to. If None, a new figure is created.
Returns:
ax (Axes): The plot axes
"""
if len(df['signedContrast'].unique()) > 4:
df2 = df.groupby(['signedContrast']).agg({'choice':'count', 'choice2':'mean'}).reset_index()
df2.rename(columns={"choice2": "fraction", "choice": "ntrials"}, inplace=True)
pars, L = psy.mle_fit_psycho(df2.transpose().values, # extract the data from the df
P_model='erf_psycho_2gammas',
parstart=np.array([df2['signedContrast'].mean(), 20., 0.05, 0.05]),
parmin=np.array([df2['signedContrast'].min(), 0., 0., 0.]),
parmax=np.array([df2['signedContrast'].max(), 100., 1, 1]))
sns.lineplot(np.arange(-100,100), psy.erf_psycho_2gammas( pars, np.arange(-100,100)), color=color, ax=ax)
# plot datapoints on top
sns.lineplot(x='signedContrast', y='choice2', err_style="bars", linewidth=0, linestyle='None', mew=0.5,
marker='.', ci=68, data=df, color=color, ax=ax)
# Reduce the clutter
ax.set_xticks([-100, -50, 0, 50, 100])
ax.set_xticklabels(['-100', '-50', '0', '50', '100'])
ax.set_yticks([0, .5, 1])
# Set the limits
ax.set_xlim([-110, 110])
ax.set_ylim([-0.03, 1.03])
ax.set_xlabel('Contrast (%)')
return ax
def plot_from_spots(spotlist,
psychoSpotData,
spots,
color,
ax,
plotType='psycho',
bs=False):
"""
Function to plot three psychometrics on a single plot, with the dots that they were
generated with. usually I'll use this with a group of left spots, a group of right spots
and a group of control spots.
Inputs:
spotlists: an array containing the spots that you want to group together to plot the
psychometric for. a n by 2 numpy array that gives [[x1,y1],[x2,y2],[xn,yn]]
psychoSpotData: a list of len(numSpots) each with a list lenght 4 containing 1, an array of
signed contrasts, 2, a list of number of presentations for that contrast, 3, a list of
arrays containing a bool for if choice was CCW (list len num contrast, array len num
presentations) 4, a list of the same form as above with the reaction times
spots: a dataframe of length num spots, column 0: laserPosX, column 1: laserPosY, column3,
count
color: the color for the line you want to plot, string eg 'b'
ax: the matplotlib axis to plot onto
plotType: defualt is psycho for plotting psychometric curves, other option is 'chrono'
"""
if plotType == 'psycho':
psycho = [[], [0 for i in range(len(psychoSpotData[0][1]))],
[np.array([])] * len(psychoSpotData[0][1])]
controlSpotPsycho = [[], [
0 for i in range(len(psychoSpotData[0][1]))
], [np.nan for i in range(len(psychoSpotData[0][1]))]]
tempPsych = [np.array([])] * len(psychoSpotData[0][1])
for i in range(len(spots)):
spotX = spots.iloc[i, [0]][0]
spotY = spots.iloc[i, [1]][0]
for spot in range(len(spotlist)):
if spotX == spotlist[spot][0] and spotY == spotlist[spot][1]:
psycho[0] = psychoSpotData[i][0]
psycho[1] = [
temp + j
for temp, j in zip(psychoSpotData[i][1], psycho[1])
]
for contrast in range(len(psycho[0])):
con = int(contrast)
tempPsych[con] = np.append(tempPsych[con],
psychoSpotData[i][2][con])
sems = []
for c in range(len(tempPsych)):
psycho[2][c] = np.nanmean(tempPsych[c])
sems.append(stats.sem(tempPsych[c]))
if bs:
## Bootstrap confidence intervals
nboots = 10
bootFits = pd.DataFrame(
columns=['threshold', 'slope', 'gamma', 'lambda'],
index=range(nboots))
bootData = [[], [0 for i in range(len(psychoSpotData[0][1]))],
[np.array([])] * len(psychoSpotData[0][1])]
bootData[0] = psycho[0]
cnt = 0
print('bootstrapping errorbars...', sep=' ', end='')
for i in range(nboots):
if not (cnt % 5):
print(int(cnt / nboots * 100),
sep=' ',
end='%,',
flush=True)
for j in range(len(tempPsych)):
bootData[2][j] = np.random.choice(
tempPsych[j],
size=int(len(tempPsych[j]) / 1.25),
replace=True)
bootData[1][j] = len(bootData[2][j])
bootData[2][j] = np.mean(bootData[2][j])
fitParams = []
fitLikes = []
for repeat in range(5):
parStart = np.array([
-5 + np.random.rand() * 10, 0 + np.random.rand() * 100,
0 + np.random.rand(), 0 + np.random.rand()
])
pars, L = mle_fit_psycho(bootData,
P_model='erf_psycho_2gammas',
parstart=np.array([0, 50, .5,
.5]),
parmin=np.array([-5, 0., 0., 0.]),
parmax=np.array([5, 100., 1, 1]),
nfits=2)
fitParams.append(pars)
fitLikes.append(L)
cnt += 1
bootFits.iloc[i] = fitParams[np.where(min(fitLikes))[0][0]]
a = .05
CIs = []
for i in bootFits.columns:
CIs.append([
np.percentile(bootFits[i], 100 - a / 2),
np.percentile(
bootFits[i],
a / 2,
)
])
else:
CIs = None
## plotting psychometrics for different cortical groups
lines = []
fitParams = []
fitLikes = []
for repeat in range(10):
parStart = np.array([
-5 + np.random.rand() * 10, 0 + np.random.rand() * 100,
0 + np.random.rand(), 0 + np.random.rand()
])
params, L = psychofit.mle_fit_psycho(
psycho,
P_model='erf_psycho_2gammas',
parstart=parStart,
parmin=np.array([-5, 0., 0., 0.]),
parmax=np.array([5, 100., 1, 1]),
nfits=25)
fitParams.append(params)
fitLikes.append(L)
# find the best params (with the lowest neg likelihood)
params = fitParams[np.where(min(fitLikes))[0][0]]
spotFits.append(params)
#plot the psychometrics
fitx = np.linspace(-1, 1, 100)
fity = psychofit.erf_psycho_2gammas(params, fitx)
line = ax.plot(fitx, fity, color=color)
lines.append(line)
ax.errorbar(psycho[0],
np.array(psycho[2]),
yerr=sems,
color=color,
marker='.',
ms=4,
ls='')
plt.ylim(-0.1, 1.1)
plt.xlim(-1.1, 1.1)
elif plotType == 'chrono':
chrono = [[], [0 for i in range(len(psychoSpotData[0][1]))],
[np.array([])] * len(psychoSpotData[0][1])]
tempChrono = [np.array([])] * len(psychoSpotData[0][1])
for i in range(len(spots)):
spotX = spots.iloc[i, [0]][0]
spotY = spots.iloc[i, [1]][0]
for spot in range(len(spotlist)):
if spotX == spotlist[spot][0] and spotY == spotlist[spot][1]:
chrono[0] = psychoSpotData[i][0]
chrono[1] = [
temp + j
for temp, j in zip(psychoSpotData[i][1], chrono[1])
]
for contrast in range(len(chrono[0])):
con = int(contrast)
tempChrono[con] = np.append(tempChrono[con],
psychoSpotData[i][3][con])
sems = []
for c in range(len(tempChrono)):
chrono[2][c] = np.nanmedian(tempChrono[c])
sems.append(stats.sem(tempChrono[c]))
ax.errorbar(chrono[0],
np.array(chrono[2]),
yerr=sems,
color=color,
marker='.',
ms=4)
ax.set_ylim(.15, .5)
# the params I use here are RT at 0 contrast, 'RT bias' which is pairwise RT left- RT right, and
# peakiness which is the ratio of max RT to the average of the two 100% RTs
p2 = (chrono[2][0] - chrono[2][-1]) + (chrono[2][1] - chrono[2][-2] +
(chrono[2][2] - chrono[2][-3]) +
(chrono[2][3] - chrono[2][-4]))
params = [
chrono[2][4], p2,
max(chrono[2]) / np.mean([chrono[2][0], chrono[2][-1]]), chrono[2]
]
CIs = None
else:
raise Exception(
"This is not a supported plot type, choose 'psycho' or 'chrono'")
return params, CIs
def plot_psychometric(x,
y,
col,
point=False,
line=True,
mark='.',
al=1,
ax=None,
**kwargs):
if not ax:
ax = plt.sca(ax[0])
# summary stats - average psychfunc over observers
df = pd.DataFrame({'signed_contrast': x, 'choice': y, 'choice2': y})
df2 = df.groupby(['signed_contrast']).agg({
'choice2': 'count',
'choice': 'mean'
}).reset_index()
df2.rename(columns={
"choice2": "ntrials",
"choice": "fraction"
},
inplace=True)
df2 = df2.groupby(['signed_contrast']).mean().reset_index()
df2 = df2[['signed_contrast', 'ntrials', 'fraction']]
# fit psychfunc
pars, L = psy.mle_fit_psycho(
df2.transpose().values, # extract the data from the df
P_model='erf_psycho_2gammas',
parstart=np.array([df2['signed_contrast'].mean(), 20., 0.05, 0.05]),
parmin=np.array([df2['signed_contrast'].min(), 5, 0., 0.]),
parmax=np.array([df2['signed_contrast'].max(), 100., 1, 1]))
if line:
# plot psychfunc
sns.lineplot(np.arange(-29, 29),
psy.erf_psycho_2gammas(pars, np.arange(-29, 29)),
color=col,
alpha=al,
ax=ax)
# plot psychfunc: -100, +100
sns.lineplot(np.arange(-37, -32),
psy.erf_psycho_2gammas(pars, np.arange(-103, -98)),
color=col,
alpha=al,
ax=ax)
sns.lineplot(np.arange(32, 37),
psy.erf_psycho_2gammas(pars, np.arange(98, 103)),
color=col,
alpha=al,
ax=ax)
# now break the x-axis
# if 100 in df.signed_contrast.values and not 50 in
# df.signed_contrast.values:
df['signed_contrast'] = df['signed_contrast'].replace(-100, -35)
df['signed_contrast'] = df['signed_contrast'].replace(100, 35)
# PLOT DATAPOINTS
if point == True:
sns.lineplot(df['signed_contrast'],
df['choice'],
err_style="bars",
linewidth=0,
linestyle='None',
mew=0.5,
marker=mark,
ci=95,
color=col,
alpha=al,
markersize=3,
ax=ax)
ax.set_xticks([-35, -25, -12.5, 0, 12.5, 25, 35])
ax.set_xticklabels(['-100', ' ', ' ', '0', ' ', ' ', '100'],
size='small',
rotation=45)
ax.set_xlim([-40, 40])
ax.set_ylim([0, 1])
ax.set_yticks([0, 0.25, 0.5, 0.75, 1])
ax.set_yticklabels(['0', ' ', '50', ' ', '100'])
def plot_psychos_from_spots(spotlists, psychoSpotData, spots, labels):
"""
Function to plot three psychometrics on a single plot, with the dots that they were
generated with. usually I'll use this with a group of left spots, a group of right spots
and a group of control spots.
Inputs:
spotlists: a list of length 3 that has the spots
"""
Psycho1 = [[],[0 for i in range(len(psychoSpotData[0][1]))],[np.nan for i in range(len(psychoSpotData[0][1]))]]
Psycho2 = [[],[0 for i in range(len(psychoSpotData[0][1]))],[np.nan for i in range(len(psychoSpotData[0][1]))]]
controlSpotPsycho = [[],[0 for i in range(len(psychoSpotData[0][1]))],[np.nan for i in range(len(psychoSpotData[0][1]))]]
for i in range(len(spots)):
spotX = spots.iloc[i, [0]][0]
spotY = spots.iloc[i, [1]][0]
for leftSpots in range(len(visLeftSpots)):
if spotX == visLeftSpots[leftSpots][0] and spotY == visLeftSpots[leftSpots][1]:
Psycho1[0] = psychoSpotData[i][0]
Psycho1[1] = [i+j for i,j in zip(psychoSpotData[i][1],Psycho1[1])]
Psycho1[2] = [np.nanmean([i,j]) for i,j in zip(psychoSpotData[i][2],Psycho1[2])]
for rightSpots in range(len(visRightSpots)):
if spotX == visRightSpots[rightSpots][0] and spotY == visRightSpots[rightSpots][1]:
Psycho2[0] = psychoSpotData[i][0]
Psycho2[1] = [i+j for i,j in zip(psychoSpotData[i][1],Psycho2[1])]
Psycho2[2] = [np.nanmean([i,j]) for i,j in zip(psychoSpotData[i][2],Psycho2[2])]
for ii in controlSpots:
if spotX == spots.iloc[ii][0] and spotY == spots.iloc[ii][1]:
controlSpotPsycho[0] = psychoSpotData[ii][0]
controlSpotPsycho[1] = [i+j for i,j in zip(psychoSpotData[ii][1],controlSpotPsycho[1])]
controlSpotPsycho[2] = [np.nanmean([i,j]) for i,j in zip(psychoSpotData[ii][2],controlSpotPsycho[2])]
## plotting psychometrics for different cortical groups
lines = []
print('Fitting psychometrics, {}/{} Done'.format(subIdx,len(data)))
fig = plt.figure()
dotColors = ['.b','.r','.g']
lineColors = ['-b','-r','-g']
plotCount = 0
for psychoPlot in [Psycho1, Psycho2, controlSpotPsycho]:
fitParams = []
fitLikes = []
for repeat in range(10):
params, L = psychofit.mle_fit_psycho(psychoPlot,
P_model='erf_psycho_2gammas',
parstart=np.array([0,20,.05,.05]),
parmin=np.array([-5, 0., 0., 0.]),
parmax=np.array([5, 100., 1, 1]),
nfits=50)
fitParams.append(params)
fitLikes.append(L)
# find the best params (with the lowest neg likelihood)
params = fitParams[np.where(min(fitLikes))[0][0]]
spotBias[i][subIdx] = params[0]
spotSlope[i][subIdx] = params[1]
spotLapseLow[i][subIdx] = params[2]
spotLapseHigh[i][subIdx] = params[3]
spotFits.append(params)
#plot the psychometrics
fitx = np.linspace(-1, 1, 100)
fity = psychofit.erf_psycho_2gammas(params, fitx)
line = plt.plot(fitx, fity, lineColors[plotCount])
lines.append(line)
plt.plot(psychoPlot[0], np.array(psychoPlot[2]), dotColors[plotCount])
plotCount+=1
## Formatting the psychometric figure
LvisLine = mpl.lines.Line2D([],[],color='blue',marker='.',label=labels[0])
RvisLine = mpl.lines.Line2D([],[],color='red',marker='.',label=labels[1])
cLine = mpl.lines.Line2D([],[],color='green',marker='.',label=labels[2])
plt.legend(handles=[LvisLine,RvisLine,cLine])
## plotting psychometrics for different cortical groups
lines = []
print('Fitting psychometrics, {}/{} Done'.format(subIdx,len(data)))
fig,axs = plt.subplots(nrows=2,ncols=2)
ax=axs[0,0]
dotColors = ['.r','.b','.g']
lineColors = ['-r','-b','-g']
plotCount = 0
for psychoPlot in [visLeftPsycho, visRightPsycho, controlSpotPsycho]:
fitParams = []
fitLikes = []
for repeat in range(10):
params, L = psychofit.mle_fit_psycho(psychoPlot,
P_model='erf_psycho_2gammas',
parstart=np.array([0,20,.05,.05]),
parmin=np.array([-5, 0., 0., 0.]),
parmax=np.array([5, 100., 1, 1]),
nfits=50)
fitParams.append(params)
fitLikes.append(L)
# find the best params (with the lowest neg likelihood)
params = fitParams[np.where(min(fitLikes))[0][0]]
spotBias[i][subIdx] = params[0]
spotSlope[i][subIdx] = params[1]
spotLapseLow[i][subIdx] = params[2]
spotLapseHigh[i][subIdx] = params[3]
spotFits.append(params)
#plot the psychometrics
fitx = np.linspace(-1, 1, 100)
fity = psychofit.erf_psycho_2gammas(params, fitx)
def model_psychometric_history(behav):
select = behav.copy()
select['t-1'] = select['trial_feedback_type'].shift(periods=1).to_numpy()
select.loc[select['choice'] == -1, 'choice'] = 0
select = select.iloc[1:,:]
#select['t-1'].fillna(0, inplace=True)
select['t-1'] = select['t-1'].astype(int)
plot_psychometric(select.loc[select['signed_contrast'],
select.loc[select['probabilityLeft'] ==i, 'signed_contrast'], palette = ['red', 'green'],
ci = 68)
sns.lineplot(data = select_50, hue = 't-1', x = select_50['signed_contrast'],
y = select_50['simulation_prob'], palette = ['red', 'green'], ci = 68)
## Functions
def run_glm(behav, example, correction = True, bias = False, cross_validation = True):
for i, nickname in enumerate(np.unique(behav['subject_nickname'])):
if np.mod(i+1, 10) == 0:
print('Loading data of subject %d of %d' % (i+1, len(
np.unique(behav['subject_nickname']))))
# Get the trials of the sessions around criterion
trials = behav.loc[behav['subject_nickname'] == nickname].copy()
if bias == True:
neutral_n = fit_psychfunc(behav[(behav['subject_nickname'] == nickname)
& (behav['probabilityLeft'] == 50)])
left_fit = fit_psychfunc(behav[(behav['subject_nickname'] == nickname)
& (behav['probabilityLeft'] == 80)])
right_fit = fit_psychfunc(behav[(behav['subject_nickname'] == nickname)
& (behav['probabilityLeft'] == 20)])
behav.loc[behav['subject_nickname'] == nickname, 'bias_n'] = \
neutral_n.loc[0, 'bias']
behav.loc[behav['subject_nickname'] == nickname, 'bias_r'] = \
right_fit.loc[0, 'bias']
behav.loc[behav['subject_nickname'] == nickname, 'bias_l'] = \
left_fit.loc[0, 'bias']
else:
fit_df = dj2pandas(trials.copy())
fit_result = fit_psychfunc(fit_df)
behav.loc[behav['subject_nickname'] == nickname, 'threshold'] = \
fit_result.loc[0, 'threshold']
## GLM
#make separate datafrme
data = trials[['index', 'trial_feedback_type',
'signed_contrast', 'choice',
'probabilityLeft']].copy()
#drop trials with odd probabilities of left
data.drop(
data['probabilityLeft'][~data['probabilityLeft'].isin([50,20,80])].index,
inplace=True)
# Rewardeded choices:
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()), 'rchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == -1), 'rchoice'] = 0
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == 1), 'rchoice'] = -1
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == 1), 'rchoice'] = 1
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()) , 'rchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == -1), 'rchoice'] = 0
# Unrewarded choices:
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()), 'uchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == -1), 'uchoice'] = -1
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == 1), 'uchoice'] = 0
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == 1), 'uchoice'] = 0
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()) , 'uchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == -1) , 'uchoice'] = 1
# Apply correction
if correction == True:
data['rchoice+1'] = \
data['rchoice'].shift(periods=-1).to_numpy()
data['uchoice+1'] = \
data['uchoice'].shift(periods=-1).to_numpy()
# Shift rewarded and unrewarded predictors by one
data.loc[:, ['rchoice', 'uchoice']] = \
data[['rchoice', 'uchoice']].shift(periods=1).to_numpy()
# Drop any nan trials
data.dropna(inplace=True)
# Make sensory predictors (no 0 predictor)
contrasts = [ 25, 100, 12.5, 6.25]
for i in contrasts:
data.loc[(data['signed_contrast'].abs() == i), i] = \
np.sign(data.loc[(data['signed_contrast'].abs() == i),
'signed_contrast'].to_numpy())
data_con = data[i].copy()
data[i] = data_con.fillna(0)
# If contrast missing break
for i in contrasts:
if np.sum(data[i]) == 0:
print('missing contrast')
missing_contrast = True
else:
missing_contrast = False
if missing_contrast == True:
continue
# Make block identity (across predictors right is positive, hence logic below)
if bias == True:
data.loc[(data['probabilityLeft'] == 50), 'block'] = 0
data.loc[(data['probabilityLeft'] == 20), 'block'] = 1
data.loc[(data['probabilityLeft'] == 80), 'block'] = -1
# Make choice in between 0 and 1 -> 1 for right and 0 for left
data.loc[data['choice'] == -1, 'choice'] = 0
# Store index
index = data['index'].copy()
# Create predictor matrix
endog = data['choice'].copy()
exog = data.copy()
exog.drop(columns=['trial_feedback_type',
'signed_contrast', 'choice',
'probabilityLeft'], inplace=True)
exog = sm.add_constant(exog)
if cross_validation == False:
X_train = exog.copy()
X_test = exog.copy()
y_train = endog.copy()
y_test = endog.copy()
else:
X_train = exog.iloc[:int(len(exog)*0.70),:].copy()
X_test = exog.iloc[int(len(endog)*0.70):,:].copy()
y_train = endog.iloc[:int(len(endog)*0.70)].copy()
y_test = endog.iloc[int(len(endog)*0.70):].copy()
# Store index
index = X_test['index'].to_numpy()
X_train.drop(columns=['index'], inplace=True)
X_test.drop(columns=['index'], inplace=True)
# Fit model
try:
logit_model = sm.Logit(y_train, X_train)
result = logit_model.fit_regularized()
# print(result.summary2())
# Store model weights
behav.loc[behav['subject_nickname'] == nickname, 'intercept'] = result.params['const'].copy()
behav.loc[behav['subject_nickname'] == nickname, 'rchoice'] = result.params['rchoice'].copy()
behav.loc[behav['subject_nickname'] == nickname, 'uchoice'] = result.params['uchoice'].copy()
mask = result.params.index.get_level_values(0)
behav.loc[behav['subject_nickname'] == nickname, '25'] = result.params[25].copy()
behav.loc[behav['subject_nickname'] == nickname, '6'] = result.params.loc[mask == 6.25][0]
behav.loc[behav['subject_nickname'] == nickname, '100'] = result.params[100].copy()
behav.loc[behav['subject_nickname'] == nickname, '12'] = result.params.loc[mask == 12.5][0]
if bias == True:
behav.loc[behav['subject_nickname'] == nickname, 'block'] = result.params['block'].copy()
if correction == True:
behav.loc[behav['subject_nickname'] == nickname, 'rchoice+1'] = result.params['rchoice+1'].copy()
behav.loc[behav['subject_nickname'] == nickname, 'uchoice+1'] = result.params['uchoice+1'].copy()
# Probabilities on test data
prob = result.predict(X_test).to_numpy()
if nickname == example:
example_model = result
# Propagate to storing dataframe
behav.loc[behav['index'].isin(index), 'simulation_prob'] = prob
except:
print('singular matrix')
return behav, example_model
def data_2_X_test (behav, correction = True, bias = True):
data = behav[['index','trial_feedback_type',
'signed_contrast', 'choice',
'probabilityLeft']].copy()
#drop trials with odd probabilities of left
data.drop(
data['probabilityLeft'][~data['probabilityLeft'].isin([50,20,80])].index,
inplace=True)
# Rewardeded choices:
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()), 'rchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == -1), 'rchoice'] = 0
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == 1), 'rchoice'] = -1
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == 1), 'rchoice'] = 1
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()) , 'rchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == -1), 'rchoice'] = 0
# Unrewarded choices:
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()), 'uchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == -1), 'uchoice'] = -1
data.loc[(data['choice'] == -1) &
(data['trial_feedback_type'] == 1), 'uchoice'] = 0
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == 1), 'uchoice'] = 0
data.loc[(data['choice'] == 0) &
(data['trial_feedback_type'].isnull()) , 'uchoice'] = 0 # NoGo trials
data.loc[(data['choice'] == 1) &
(data['trial_feedback_type'] == -1) , 'uchoice'] = 1
# Apply correction
if correction == True:
data['rchoice+1'] = \
data['rchoice'].shift(periods=-1).to_numpy()
data['uchoice+1'] = \
data['uchoice'].shift(periods=-1).to_numpy()
# Shift rewarded and unrewarded predictors by one
data.loc[:, ['rchoice', 'uchoice']] = \
data[['rchoice', 'uchoice']].shift(periods=1).to_numpy()
# Drop any nan trials
data.dropna(inplace=True)
# Make sensory predictors (no 0 predictor)
contrasts = [ 25, 100, 12.5, 6.25]
for i in contrasts:
data.loc[(data['signed_contrast'].abs() == i), i] = \
np.sign(data.loc[(data['signed_contrast'].abs() == i),
'signed_contrast'].to_numpy())
data_con = data[i].copy()
data[i] = data_con.fillna(0)
# Make block identity (across predictors right is positive, hence logic below)
if bias == True:
data.loc[(data['probabilityLeft'] == 50), 'block'] = 0
data.loc[(data['probabilityLeft'] == 20), 'block'] = 1
data.loc[(data['probabilityLeft'] == 80), 'block'] = -1
# Make choice in between 0 and 1 -> 1 for right and 0 for left
data.loc[data['choice'] == -1, 'choice'] = 0
index = data['index'].copy()
# Create predictor matrix
endog = data['choice'].copy()
exog = data.copy()
exog.drop(columns=['index', 'trial_feedback_type',
'signed_contrast', 'choice',
'probabilityLeft'], inplace=True)
exog = sm.add_constant(exog)
return exog, index
def plot_psychometric(x, y, col, point = False, mark = 'o', al =1):
# summary stats - average psychfunc over observers
df = pd.DataFrame({'signed_contrast': x, 'choice': y,
'choice2': y})
df2 = df.groupby(['signed_contrast']).agg(
{'choice2': 'count', 'choice': 'mean'}).reset_index()
df2.rename(columns={"choice2": "ntrials",
"choice": "fraction"}, inplace=True)
df2 = df2.groupby(['signed_contrast']).mean().reset_index()
df2 = df2[['signed_contrast', 'ntrials', 'fraction']]
# fit psychfunc
pars, L = psy.mle_fit_psycho(df2.transpose().values, # extract the data from the df
P_model='erf_psycho_2gammas',
parstart=np.array(
[df2['signed_contrast'].mean(), 20., 0.05, 0.05]),
parmin=np.array(
[df2['signed_contrast'].min(), 5, 0., 0.]),
parmax=np.array([df2['signed_contrast'].max(), 100., 1, 1]))
# plot psychfunc
g = sns.lineplot(np.arange(-29, 29),
psy.erf_psycho_2gammas(pars, np.arange(-29, 29)), color = col,
alpha = al)
# plot psychfunc: -100, +100
sns.lineplot(np.arange(-37, -32),
psy.erf_psycho_2gammas(pars, np.arange(-103, -98)), color = col,
alpha = al)
sns.lineplot(np.arange(32, 37),
psy.erf_psycho_2gammas(pars, np.arange(98, 103)), color = col,
alpha = al)
# now break the x-axis
# if 100 in df.signed_contrast.values and not 50 in
# df.signed_contrast.values:
df['signed_contrast'] = df['signed_contrast'].replace(-100, -35)
df['signed_contrast'] = df['signed_contrast'].replace(100, 35)
if point == True:
sns.lineplot(df['signed_contrast'], df['choice'], err_style="bars",
linewidth=0, linestyle='None', mew=0.5,
marker=mark, ci=68, color = col, alpha = al)
g.set_xticks([-35, -25, -12.5, 0, 12.5, 25, 35])
g.set_xticklabels(['-100', '-25', '-12.5', '0', '12.5', '25', '100'],
size='small', rotation=45)
g.set_xlim([-40, 40])
g.set_ylim([0, 1])
g.set_yticks([0, 0.25, 0.5, 0.75, 1])
g.set_yticklabels(['0', '25', '50', '75', '100'])
# FUNCTION UNDER DEVELOPMENT
def updating:
select = behav.copy()
select['signed_contrast-1'] = select['signed_contrast'].shift(periods=1).to_numpy()
select['signed_contrast+1'] = select['signed_contrast'].shift(periods=-1).to_numpy()
select['t-1'] = select['trial_feedback_type'].shift(periods=1).to_numpy()
select['t+1'] = select['trial_feedback_type'].shift(periods=-1).to_numpy()
select = select.iloc[1:-1,:] # First and last trial will have nan for history
select['t-1'] = select['t-1'].astype(int)
select['simulation_prob'] = select['simulation_prob']*100
#select = select.loc[select['signed_contrast-1'] >= 0]
for mouse in select['subject_nickname'].unique():
for c in select['signed_contrast-1'].unique():
for r in select['t-1'].unique():
sub_select = select.loc[(select['signed_contrast-1'] == c) &
(select['t-1'] == r) & (select['subject_nickname'] == mouse)]
fit_result = fit_psychfunc(sub_select)
select.loc[select['subject_nickname'] == nickname, 'updating'] = \
fit_result['bias'][0]
for c in select['signed_contrast+1'].unique():
for r in select['t+1'].unique():
sub_select = select.loc[(select['signed_contrast+1'] == c) &
(select['t+1'] == r) & (select['subject_nickname'] == mouse)]
fit_result = fit_psychfunc(sub_select)
select.loc[select['subject_nickname'] == nickname, 'updating_correction'] = \
fit_result['bias'][0]
sns.lineplot(data = select, hue = 't-1', x = select['signed_contrast-1'],
y = select['simulation_prob'], ci = 68)
sns.lineplot(data = select, hue = 't-1', x = select['signed_contrast-1'],
y = select[select['probabilityLeft'] == 80],'choice'] -
select.loc[select['probabilityLeft'] == 20 ,'choice'])
def plot_psychometric(x, y, subj, **kwargs):
# summary stats - average psychfunc over observers
df = pd.DataFrame({
'signed_contrast': x,
'choice': y,
'choice2': y,
'subject_nickname': subj
})
df2 = df.groupby(['signed_contrast', 'subject_nickname']).agg({
'choice2':
'count',
'choice':
'mean'
}).reset_index()
df2.rename(columns={
"choice2": "ntrials",
"choice": "fraction"
},
inplace=True)
df2 = df2.groupby(['signed_contrast']).mean().reset_index()
df2 = df2[['signed_contrast', 'ntrials', 'fraction']]
# fit psychfunc
pars, L = psy.mle_fit_psycho(
df2.transpose().values, # extract the data from the df
P_model='erf_psycho_2gammas',
parstart=np.array([df2['signed_contrast'].mean(), 20., 0.05, 0.05]),
parmin=np.array([df2['signed_contrast'].min(), 0., 0., 0.]),
parmax=np.array([df2['signed_contrast'].max(), 100., 0.5, 0.5]))
# plot psychfunc
sns.lineplot(np.arange(-29, 29),
psy.erf_psycho_2gammas(pars, np.arange(-29, 29)), **kwargs)
# plot psychfunc: -100
sns.lineplot(np.arange(-37, -32),
psy.erf_psycho_2gammas(pars, np.arange(-103, -98)), **kwargs)
sns.lineplot(np.arange(32, 37),
psy.erf_psycho_2gammas(pars, np.arange(98, 103)), **kwargs)
# now break the x-axis
# if 100 in df.signed_contrast.values and not 50 in df.signed_contrast.values:
df['signed_contrast'] = df['signed_contrast'].replace(-100, -35)
df['signed_contrast'] = df['signed_contrast'].replace(100, 35)
df3 = df.groupby(['signed_contrast', 'subject_nickname']).agg({
'choice2':
'count',
'choice':
'mean'
}).reset_index()
# plot datapoints with errorbars on top
g = sns.lineplot(df3['signed_contrast'],
df3['choice'],
err_style="bars",
linewidth=0,
linestyle='None',
mew=0.5,
marker='o',
ci=68,
**kwargs)
g.set_yticks([0, 0.25, 0.5, 0.75, 1])
# # ADD TEXT WITH THE PSYCHOMETRIC FUNCTION PARAMETERS
# if len(df['subject_nickname'].unique()) == 1:
# try:
# # add text with parameters into the plot
# if kwargs['label'] == '50':
# ypos = 0.5
# # ADD PSYCHOMETRIC FUNCTION PARAMS
# plt.text(-35, ypos, r'$\mu\/ %.2f,\/ \sigma\/ %.2f,$'%(pars[0], pars[1]) + '\n' + r'$\gamma \/%.2f,\/ \lambda\/ %.2f$'%(pars[2], pars[3]),
# fontweight='normal', fontsize=5, color=kwargs['color'])
# elif kwargs['label'] == '20':
# ypos = 0.3
# # ADD PSYCHOMETRIC FUNCTION PARAMS
# plt.text(-35, ypos, r'$\mu\/ %.2f,\/ \sigma\/ %.2f,$'%(pars[0], pars[1]) + '\n' + r'$\gamma \/%.2f,\/ \lambda\/ %.2f$'%(pars[2], pars[3]),
# fontweight='normal', fontsize=5, color=kwargs['color'])
# elif kwargs['label'] == '80':
# ypos = 0.7
# # ADD PSYCHOMETRIC FUNCTION PARAMS
# plt.text(-35, ypos, r'$\mu\/ %.2f,\/ \sigma\/ %.2f,$'%(pars[0], pars[1]) + '\n' + r'$\gamma \/%.2f,\/ \lambda\/ %.2f$'%(pars[2], pars[3]),
# fontweight='normal', fontsize=5, color=kwargs['color'])
# except: # when there is no label
# # pass
# ypos = 0.5
# # ADD PSYCHOMETRIC FUNCTION PARAMS
# plt.text(-35, ypos, r'$\mu\/ %.2f,\/ \sigma\/ %.2f,$'%(pars[0], pars[1]) + '\n' + r'$\gamma \/%.2f,\/ \lambda\/ %.2f$'%(pars[2], pars[3]),
# fontweight='normal', fontsize=8, color=kwargs['color'])
# # plt.text(12, 0.1, '1 mouse', fontsize=10, color='k')
# print the number of mice
if df['subject_nickname'].nunique() == 1:
plt.text(12, 0.1, '1 mouse', fontsize=10, color='k')
else:
plt.text(12,
0.1,
'%d mice' % (df['subject_nickname'].nunique()),
fontsize=10,
color='k')
#if brokenXaxis:
g.set_xticks([-35, -25, -12.5, 0, 12.5, 25, 35])
g.set_xticklabels(['-100', '-25', '-12.5', '0', '12.5', '25', '100'],
size='small',
rotation=45)
g.set_xlim([-40, 40])
g.set_ylim([0, 1])
g.set_yticks([0, 0.25, 0.5, 0.75, 1])
g.set_yticklabels(['0', '25', '50', '75', '100'])
def plot_psychometric(x, y, subj, **kwargs):
# summary stats - average psychfunc over observers
df = pd.DataFrame({'signed_contrast': x, 'choice': y,
'choice2': y, 'subject_nickname': subj})
df2 = df.groupby(['signed_contrast', 'subject_nickname']).agg(
{'choice2': 'count', 'choice': 'mean'}).reset_index()
df2.rename(columns={"choice2": "ntrials",
"choice": "fraction"}, inplace=True)
df2 = df2.groupby(['signed_contrast']).mean().reset_index()
df2 = df2[['signed_contrast', 'ntrials', 'fraction']]
# only 'break' the x-axis and remove 50% contrast when 0% is present
# print(df2.signed_contrast.unique())
if 0. in df2.signed_contrast.values:
brokenXaxis = True
else:
brokenXaxis = False
# fit psychfunc
pars, L = psy.mle_fit_psycho(df2.transpose().values, # extract the data from the df
P_model='erf_psycho_2gammas',
parstart=np.array(
[0, 20., 0.05, 0.05]),
parmin=np.array(
[df2['signed_contrast'].min(), 5, 0., 0.]),
parmax=np.array([df2['signed_contrast'].max(), 40., 1, 1]))
if brokenXaxis:
# plot psychfunc
g = sns.lineplot(np.arange(-29, 29),
psy.erf_psycho_2gammas(pars, np.arange(-29, 29)), **kwargs)
# plot psychfunc: -100, +100
sns.lineplot(np.arange(-37, -32),
psy.erf_psycho_2gammas(pars, np.arange(-103, -98)), **kwargs)
sns.lineplot(np.arange(32, 37),
psy.erf_psycho_2gammas(pars, np.arange(98, 103)), **kwargs)
# now break the x-axis
# if 100 in df.signed_contrast.values and not 50 in
# df.signed_contrast.values:
df['signed_contrast'] = df['signed_contrast'].replace(-100, -35)
df['signed_contrast'] = df['signed_contrast'].replace(100, 35)
else:
# plot psychfunc
g = sns.lineplot(np.arange(-103, 103),
psy.erf_psycho_2gammas(pars, np.arange(-103, 103)), **kwargs)
df3 = df.groupby(['signed_contrast', 'subject_nickname']).agg(
{'choice2': 'count', 'choice': 'mean'}).reset_index()
# plot datapoints with errorbars on top
if df['subject_nickname'].nunique() > 1:
sns.lineplot(df3['signed_contrast'], df3['choice'], err_style="bars",
linewidth=0, linestyle='None', mew=0.5,
marker='o', ci=68, **kwargs)
if brokenXaxis:
g.set_xticks([-35, -25, -12.5, 0, 12.5, 25, 35])
g.set_xticklabels(['-100', '-25', '-12.5', '0', '12.5', '25', '100'],
size='small', rotation=60)
g.set_xlim([-40, 40])
else:
g.set_xticks([-100, -50, 0, 50, 100])
g.set_xticklabels(['-100', '-50', '0', '50', '100'],
size='small', rotation=60)
g.set_xlim([-110, 110])
g.set_ylim([0, 1.02])
g.set_yticks([0, 0.25, 0.5, 0.75, 1])
g.set_yticklabels(['0', '25', '50', '75', '100'])
