def __init__(self,
signal,
freq,
file=False,
filename=None,
show_plot=False):
self._y = signal
self._freq = freq
self._derivative = self._get_derivative(signal)
if file:
self.data = nk.eda_process(signal, sampling_rate=freq)
if show_plot:
nk.plot_events_in_signal(nk.z_score(self.data["df"]),
self.data["EDA"]["SCR_Peaks_Indexes"])
print(self.data["EDA"]["SCR_Peaks_Indexes"])
plt.show()
self.save_to_file(filename, self.data)
else:
self.data = self.read_from_file(filename)
def test_z_score():
raw_scores = [1, 1, 5, 2, 1]
z = nk.z_score(raw_scores)
assert int(z.values[3]) == 0
)
# Plot it
df.plot()
plt.title('All the data ploted')
plt.savefig('graphs/alldata')
# Process the signals
bio = nk.bio_process(ecg=df["ECG"],
rsp=df["RSP"],
eda=df["EDA"],
add=df["Photosensor"],
sampling_rate=100)
# Plot the processed dataframe, normalizing all variables for viewing purpose
nk.z_score(bio["df"]).plot()
plt.title('Signals plotted')
plt.savefig('graphs/signalsplot')
# Get average quality
print("AVG quality:")
print(bio["ECG"]["Average_Signal_Quality"])
# Plot all the heart beats
pd.DataFrame(bio["ECG"]["Cardiac_Cycles"]).plot(legend=False)
plt.title('Heart beats')
plt.savefig('graphs/hbeats')
# A large number of HRV indices can be found
print("HRV indices: ")
print(bio["ECG"]["HRV"])
# ## Extracting Features # To understand the ECG and RESP signals, we must extract some features out of them. I am using a package called NeuroKit that does exactly that. It extracts information from the ECG and respiration signals to obtain simple values like heart rate, and heart beat positions. It also extracts some more complex values like Respiratory Sinus Arrhythmia, which is a naturally occurring variation in heart rate that occurs during a breathing cycle (inhalation and exhalation). This variation is directly affected by the autonomic nervous system, and thus, by measuring this variation of the heart rate during stress and non-stress conditions, we can capture useful information in classifying the two states. # # To see which data is most useful, I have plotted it for all the subjects and analyzed all the features, as well as gotten their standard deviations and other metrics for analysis. For brevity, I have kept some findings for one subject and have only kept a few example plots that make it easy to visualize the type of things I was looking for in the features. # In[353]: # Process Example ECG Signals for Visual Analysis i = 1 bio = nk.bio_process(ecg=df_respiban[i][:]['ECG'],rsp=df_respiban[i][:]['RESP'], ecg_quality_model=None, ecg_filter_frequency=[1,35], sampling_rate=100) # Plot the processed dataframe, normalizing all variables for viewing purpose nk.z_score(bio["df"][1000:-1000]).plot() plt.show() # In[551]: nk.z_score(bio["df"][["ECG_Filtered", "RSP_Filtered", "RSA"]])[1000:2500].plot() plt.show() # In[552]: rsa_values = [0,0] for i in range(1,2):
def eda_ERP(epoch, event_length, sampling_rate=1000, window_post=4):
"""
Extract event-related EDA and Skin Conductance Response (SCR).
Parameters
----------
epoch : pandas.DataFrame
An epoch contains in the epochs dict returned by :function:`nk.create_epochs()` on dataframe returned by :function:`nk.bio_process()`. Index must range from -4s to +4s (relatively to event onset and end).
event_length : int
In seconds.
sampling_rate : int
Sampling rate (samples/second).
window_post : float
Post-stimulus window size (in seconds) to include eventual responses (usually 3 or 4).
Returns
----------
SCR : dict
Event-locked SCR response features.
Example
----------
>>> import neurokit as nk
>>> bio = nk.bio_process(eda=df["EDA"], add=df["Photosensor"])
>>> df = bio["df"]
>>> events = nk.find_events(df["Photosensor"], cut="lower")
>>> epochs = nk.create_epochs(df, events["onsets"], duration=events["durations"]+8000, onset=-4000)
>>> for epoch in epochs:
>>> SCR = nk.eda_eventlocked_response(epoch, event_length=4000)
Notes
----------
*Details*
- **Looking for help**: *Experimental*: respiration artifacts correction needs to be implemented.
- **EDA_Peak**: Max of EDA (in a window starting 1s after the stim onset) minus baseline.
- **SCR_Amplitude**: Peak of SCR. If no SCR, returns NA.
- **SCR_Magnitude**: Peak of SCR. If no SCR, returns 0.
- **SCR_Amplitude_Log**: log of 1+amplitude.
- **SCR_Magnitude_Log**: log of 1+magnitude.
- **SCR_PeakTime**: Time of peak.
- **SCR_Latency**: Time between stim onset and SCR onset.
- **SCR_RiseTime**: Time between SCR onset and peak.
- **SCR_Strength**: *Experimental*: peak divided by latency.
*Authors*
- Dominique Makowski (https://github.com/DominiqueMakowski)
*Dependencies*
- numpy
- pandas
*See Also*
- https://www.biopac.com/wp-content/uploads/EDA-SCR-Analysis.pdf
References
-----------
- Schneider, R., Schmidt, S., Binder, M., Schäfer, F., & Walach, H. (2003). Respiration-related artifacts in EDA recordings: introducing a standardized method to overcome multiple interpretations. Psychological reports, 93(3), 907-920.
"""
# Initialization
event_length = event_length/sampling_rate*1000
SCR = {}
# Sanity check
if epoch.index[-1]/sampling_rate*1000-event_length < 1000:
print("NeuroKit Warning: eda_eventlocked_response(): your epoch only lasts for about %.2f s post stimulus. You might lose some SCRs." %((epoch.index[-1]/sampling_rate*1000-event_length)/1000))
if epoch.index[0]/sampling_rate*1000 > -3000:
print("NeuroKit Warning: eda_eventlocked_response(): your epoch only starts %.2f s before the stimulus. Might induce some errors in artifacts correction." %((epoch.index[0]/sampling_rate*1000)/1000))
# EDA Based
# =================
baseline = epoch["EDA_Filtered"].ix[0]
eda_peak = epoch["EDA_Filtered"].ix[sampling_rate:event_length+window_post*sampling_rate].max()
SCR["EDA_Peak"] = eda_peak - baseline
# SCR Based
# =================
# Very Basic Model
# SCR["SCR_Amplitude_Basic"] = epoch["SCR_Peaks"].ix[100:event_length+4*sampling_rate].max()
# if np.isnan(SCR["SCR_Amplitude_Basic"]):
# SCR["SCR_Magnitude_Basic"] = 0
# else:
# SCR["SCR_Magnitude_Basic"] = SCR["SCR_Amplitude_Basic"]
# Model
peak_onset = epoch["SCR_Onsets"].ix[0:event_length].idxmax()
if pd.isnull(peak_onset) is False:
SCR["SCR_Amplitude"] = epoch["SCR_Peaks"].ix[peak_onset:event_length+window_post*sampling_rate].max()
peak_loc = epoch["SCR_Peaks"].ix[peak_onset:event_length+window_post*sampling_rate].idxmax()
SCR["SCR_Magnitude"] = SCR["SCR_Amplitude"]
if pd.isnull(SCR["SCR_Amplitude"]):
SCR["SCR_Magnitude"] = 0
else:
SCR["SCR_Amplitude"] = np.nan
SCR["SCR_Magnitude"] = 0
# Log
SCR["SCR_Amplitude_Log"] = np.log(1+SCR["SCR_Amplitude"])
SCR["SCR_Magnitude_Log"] = np.log(1+SCR["SCR_Magnitude"])
# Latency and Rise time
if np.isfinite(SCR["SCR_Amplitude"]):
peak_onset = epoch["SCR_Onsets"].ix[0:peak_loc].idxmax()
SCR["SCR_PeakTime"] = peak_loc/sampling_rate*1000
SCR["SCR_Latency"] = peak_onset/sampling_rate*1000
SCR["SCR_RiseTime"] = (peak_loc - peak_onset)/sampling_rate*1000
else:
SCR["SCR_PeakTime"] = np.nan
SCR["SCR_Latency"] = np.nan
SCR["SCR_RiseTime"] = np.nan
SCR["SCR_Strength"] = SCR["SCR_Magnitude"]/(SCR["SCR_Latency"]/1000)
# RSP Corrected
# This needs to be done!!
if "RSP_Filtered" in epoch.columns:
# granger = statsmodels.tsa.stattools.grangercausalitytests(epoch[["EDA_Filtered", "RSP_Filtered"]], 10)
RSP_z = nk.z_score(epoch["RSP_Filtered"])
RSP_peak = RSP_z.ix[:0].max()
if np.isnan(RSP_peak[0]) and RSP_peak[0] > 1.96:
SCR["SCR_Amplitude_RSP_Corrected"] = SCR["SCR_Amplitude"]/(RSP_peak-0.96)
SCR["SCR_Magnitude_RSP_Corrected"] = SCR["SCR_Magnitude"]/(RSP_peak-0.96)
else:
SCR["SCR_Amplitude_RSP_Corrected"] = SCR["SCR_Amplitude"]
SCR["SCR_Magnitude_RSP_Corrected"] = SCR["SCR_Magnitude"]
return(SCR)
#
#df = pd.concat(All, axis=0, ignore_index=True)
#df.to_csv("cardiac_cycles.csv", index=False)
df = pd.DataFrame.from_csv("cardiac_cycles.csv")
#==============================================================================
# Preprocessing
#==============================================================================
# Remove lines with NA
df = df.dropna(axis=1)
# Z score
y = df.pop("Lead")
df = nk.z_score(df.T).T
X = np.array(df)
X_train, X_test, y_train, y_test = sklearn.model_selection.train_test_split(X, y, test_size=0.33, random_state=666)
##==============================================================================
## Fit model
##==============================================================================
model = sklearn.neural_network.MLPClassifier()
#
model.fit(X_train, y_train)
def test_z_score(self):
raw_scores = [1, 1, 5, 2, 1]
z = nk.z_score(raw_scores)
self.assertEqual(int(z.values[3]), 0)
rri_new = scipy.interpolate.splev(x=np.arange(0, beat_time[-1], 1/sampling_rate), tck=spline, der=0)
# plotting
plt.figure(figsize=(20, 3), dpi=100)
plt.scatter(np.arange(rri_new.shape[0]), rri_new,s=1)
plt.grid(True)
plt.title('RRI indexed per beat')
plt.ylabel('RR interval duration (sec)')
plt.xlabel('#beat')
plt.tight_layout()
plt.show()
plt.figure(figsize=(20, 3), dpi=100)
plt.plot(rri,'b-',linewidth=0.5)
plt.grid(True)
plt.title('Resampled RRI indexed per time')
plt.ylabel('RR interval duration (sec)')
plt.xlabel('Time (sec)')
plt.tight_layout()
plt.show()
df=df["df"][["ECG_Filtered", "Heart_Rate"]]
df["ECG_HRV"] = np.nan
df["ECG_HRV"].ix[rpeaks[0]+1:rpeaks[0]+len(rri_new)] = rri_new
nk.z_score(df).plot()