Errors are an option... 

# Mandatory Tasks

## 2 

Basic: 

![image0.png](image0.png)

Scaled by divison max() for every individual frequency band:

![image1.png](image1.png)

## 3

without preparing the data via /= std: 

![image2.png](image2.png)

with equalizing the power via /= std (obviously not the best idea in this case):

![image3.png](image3.png)

## 4

Phase Coherence

![image4.png](image4.png)


![image5.png](image5.png)

Spectral Coherence

![image6.png](image6.png)


![image7.png](image7.png)

## 5

Phase Coherence

![image10.png](image10.png)


![image11.png](image11.png)


Spectral Coherence

![image8.png](image8.png)


![image9.png](image9.png)

## 6

{: .topic-optional}
Don't normalize the time series! 

i.e. don't do something like this: 

```python
data -= data.mean(axis=1, keepdims=True)
data /= data.std(axis=1, keepdims=True)
```
Otherwise you will not classify anything.

![image12.png](image12.png)

Scaled by divison max() for every individual frequency band (Bad times happen):

![image13.png](image13.png)

## 7

{: .topic-optional}
Don't normalize the time series! 

i.e. don't do something like this: 

```python
data -= data.mean(axis=1, keepdims=True)
data /= data.std(axis=1, keepdims=True)
```
Otherwise you will not classify anything.

![image14.png](image14.png)