---
tags: Python Workshop 沈煒翔
---

# Lesson 8: Plotting

## Line plot

Matplotlib is a library commonly used for plotting figures in Python.

import matplotlib.pyplot as plt
import numpy as np


x = np.linspace(0, 2, 100)  # Sample data.

plt.figure()
plt.plot(x, x, label='linear')  # Plot some data on the (implicit) axes.
plt.plot(x, x**2, label='quadratic')  # etc.
plt.plot(x, x**3, label='cubic')
plt.xlabel('x label')
plt.ylabel('y label')
plt.title("Simple Plot")
plt.legend()

In some environments, you have to call ```plt.show()``` at the end of the script to show the figures.


You can control some properties of the line plot.

plt.figure()
plt.plot(x, x, 'r.', label='linear') # color + line style
plt.plot(x, x**2, '.-', label='quadratic') # line style


## Scatter plot

We can plot a 2D scatter plot.

x = np.random.normal(loc=0, scale=1, size=200)
y = np.random.normal(loc=0, scale=1, size=200)
x2 = np.random.normal(loc=3, scale=2, size=200)
y2 = np.random.normal(loc=2, scale=2, size=200)

plt.figure()
plt.scatter(x, y)
plt.scatter(x2, y2)
plt.legend(['a', 'b'])


## Histogram

We can plot a 1D histogram

x = np.random.normal(loc=10, scale=3, size=5000)

plt.figure()
plt.hist(x, bins=50)


## Bar chart

We can plot a bar chart to display multiple 1D data.

people = ('Tom', 'Dick', 'Harry', 'Slim', 'Jim')
y_pos = np.arange(len(people))
performance = 3 + 10 * np.random.rand(len(people))
error = np.random.rand(len(people))

plt.figure()
plt.barh(y_pos, performance, xerr=error, align='center')
plt.yticks(ticks=y_pos, labels=people)


### Exercise

Plot a line ```y = x^2 + 3x -5``` and plot another line in points using the same equation but with some errors.

x = np.linspace(-5, 5, 30)


### Exercise

Plot three 2D normal distribution.


Input a random point (e.g. (0,0)), use the nearest neighbor method to find which distribution it belongs to.

sample_point = (0, 0)

## Data processing

Assume we have a 1D time series that is noisy. We can apply moving average smoothing to eliminate some noise.

x = np.linspace(-5, 5, 100)
y = x**2 + 3*x - 5 + 2*np.random.randn(100)

plt.figure()
plt.plot(x, y, '.-')

y_smoothed = np.zeros(x.shape)

for i in range(5, len(x)-5):
    y_smoothed[i] = np.mean(y[i-5:i+5])

y_smoothed[:5] = np.isnan
y_smoothed[-5:] = np.isnan
	
plt.figure()
plt.plot(x, y, '.-')
plt.plot(x, y_smoothed, '.-')


### Exercise

Assume we have a 1D time series, but some points are loss and record as nan.

Correct the nan using the average of the previous and the next value.

x = np.linspace(-5, 5, 100)
y = x**2 + 3*x - 5

y_drop = y
for i in range(100):
    if np.random.random() < 0.05:
        y_drop[i] = np.nan

plt.figure()
plt.plot(x, y_drop, '.-')

# YOUR CORRECTION
# HINT: use np.isnan() to detect nan values!

# YOUR PLOTTING

Solution:

# YOUR CORRECTION
# HINT: use np.isnan() to detect nan values!
y_corrected = np.zeros(y_drop.shape)

for i in range(len(y_drop)):
    if np.isnan(y_drop[i]):
        # reconstruct
        y_corrected[i] = (y_drop[i-1] + y_drop[i+1])/2
    else:
        y_corrected[i] = y_drop[i]

# YOUR PLOTTING
plt.figure()
plt.plot(x, y_corrected, '.-')

## Optimization

Assume we observe a data and want to model it with a ```y=ax+b``` system.

x = np.linspace(-5, 5, 100)
y = 3*x - 5 + np.random.randn(100)

plt.figure()
plt.plot(x, y)


We can solve the optimal (a, b) using linear sweeping.

optimal_error = 1e20
optimal_a = 0
optimal_b = 0

for a in np.arange(-10, 10, 0.1):
    for b in np.arange(-10, 10, 0.1):
        y_pred = a*x + b
        error = np.mean((y_pred - y)**2)
        
        if error < optimal_error:
            optimal_a = a
            optimal_b = b
            optimal_error = error

print(optimal_a, optimal_b)            

Then we can plot it.

y_pred = optimal_a*x + optimal_b

plt.figure()
plt.plot(x, y)
plt.plot(x, y_pred)


Linear sweeping is the most naive approach and takes a lot of time. You should learn more advance optimization techniques (e.g. gradient descent) later in the course.

### Exercise

Assume there is a ```y=ax^2+bx+c``` system. Follow the above methods.

1. Generate the 1D data (with noise)
2. Use linear sweeping to model it with a ```y=ax+b``` system
3. Plot (line plot) both data (observation/prediction)