Logistic regression with one variable (High melting explosives)

Contents

Hide code cell source
MAKE_BOOK_FIGURES=Trueimport numpy as npimport scipy.stats as stimport matplotlib as mplimport matplotlib.pyplot as plt%matplotlib inlineimport matplotlib_inlinematplotlib_inline.backend_inline.set_matplotlib_formats('svg')import seaborn as snssns.set_context("paper")sns.set_style("ticks")def set_book_style():    plt.style.use('seaborn-v0_8-white')     sns.set_style("ticks")    sns.set_palette("deep")    mpl.rcParams.update({        # Font settings        'font.family': 'serif',  # For academic publishing        'font.size': 8,  # As requested, 10pt font        'axes.labelsize': 8,        'axes.titlesize': 8,        'xtick.labelsize': 7,  # Slightly smaller for better readability        'ytick.labelsize': 7,        'legend.fontsize': 7,                # Line and marker settings for consistency        'axes.linewidth': 0.5,        'grid.linewidth': 0.5,        'lines.linewidth': 1.0,        'lines.markersize': 4,                # Layout to prevent clipped labels        'figure.constrained_layout.use': True,                # Default DPI (will override when saving)        'figure.dpi': 600,        'savefig.dpi': 600,                # Despine - remove top and right spines        'axes.spines.top': False,        'axes.spines.right': False,                # Remove legend frame        'legend.frameon': False,                # Additional trim settings        'figure.autolayout': True,  # Alternative to constrained_layout        'savefig.bbox': 'tight',    # Trim when saving        'savefig.pad_inches': 0.1   # Small padding to ensure nothing gets cut off    })def set_notebook_style():    plt.style.use('seaborn-v0_8-white')    sns.set_style("ticks")    sns.set_palette("deep")    mpl.rcParams.update({        # Font settings - using default sizes        'font.family': 'serif',        'axes.labelsize': 10,        'axes.titlesize': 10,        'xtick.labelsize': 9,        'ytick.labelsize': 9,        'legend.fontsize': 9,                # Line and marker settings        'axes.linewidth': 0.5,        'grid.linewidth': 0.5,        'lines.linewidth': 1.0,        'lines.markersize': 4,                # Layout settings        'figure.constrained_layout.use': True,                # Remove only top and right spines        'axes.spines.top': False,        'axes.spines.right': False,                # Remove legend frame        'legend.frameon': False,                # Additional settings        'figure.autolayout': True,        'savefig.bbox': 'tight',        'savefig.pad_inches': 0.1    })def save_for_book(fig, filename, is_vector=True, **kwargs):    """    Save a figure with book-optimized settings.        Parameters:    -----------    fig : matplotlib figure        The figure to save    filename : str        Filename without extension    is_vector : bool        If True, saves as vector at 1000 dpi. If False, saves as raster at 600 dpi.    **kwargs : dict        Additional kwargs to pass to savefig    """        # Set appropriate DPI and format based on figure type    if is_vector:        dpi = 1000        ext = '.pdf'    else:        dpi = 600        ext = '.tif'        # Save the figure with book settings    fig.savefig(f"{filename}{ext}", dpi=dpi, **kwargs)def make_full_width_fig():    return plt.subplots(figsize=(4.7, 2.9), constrained_layout=True)def make_half_width_fig():    return plt.subplots(figsize=(2.35, 1.45), constrained_layout=True)if MAKE_BOOK_FIGURES:    set_book_style()else:    set_notebook_style()make_full_width_fig = make_full_width_fig if MAKE_BOOK_FIGURES else lambda: plt.subplots()make_half_width_fig = make_half_width_fig if MAKE_BOOK_FIGURES else lambda: plt.subplots()

Logistic regression with one variable (High melting explosives)#

High Melting Explosives (HMX) have applications as detonators of nuclear weapons and as solid rocket propellants. We will use logistic regression to build the probability that a specific HMX block explodes when dropped from a given height. To this end, we will use data from a 1987 Los Alamos Report (L. Smith, “Los Alamos National Laboratory explosives orientation course: Sensitivity and sensitivity tests to impact, friction, spark and shock,” Los Alamos National Lab, NM (USA), Tech. Rep., 1987). Let’s download the raw data and load them. We will use the Python Data Analysis Library:

url = "https://github.com/PredictiveScienceLab/data-analytics-se/raw/master/lecturebook/data/hmx_data.csv"
!curl -O $url

import pandas as pd
data = pd.read_csv('hmx_data.csv')

Each row of the data is a different experiment. There are two columns:

  • The first column is Height: From what height (in cm) the specimen was dropped?

  • The second column is Result: Did the specimen explode (E) or not (N)?

Here is how to see the raw data:

data
Hide code cell output
Height Result
0 40.5 E
1 40.5 E
2 40.5 E
3 40.5 E
4 40.5 E
5 40.5 E
6 40.5 E
7 40.5 E
8 40.5 E
9 40.5 E
10 36.0 E
11 36.0 N
12 36.0 E
13 36.0 E
14 36.0 E
15 36.0 E
16 36.0 N
17 36.0 E
18 36.0 E
19 36.0 E
20 32.0 E
21 32.0 E
22 32.0 N
23 32.0 E
24 32.0 E
25 32.0 E
26 32.0 N
27 32.0 E
28 32.0 N
29 32.0 E
30 28.5 N
31 28.5 E
32 28.5 N
33 28.5 N
34 28.5 E
35 28.5 N
36 28.5 N
37 28.5 N
38 28.5 E
39 28.5 N
40 25.5 N
41 25.5 N
42 25.5 N
43 25.5 N
44 25.5 N
45 25.5 N
46 25.5 E
47 25.5 N
48 25.5 N
49 25.5 N
50 22.5 N
51 22.5 N
52 22.5 N
53 22.5 N
54 22.5 N
55 22.5 N
56 22.5 N
57 22.5 N
58 22.5 N
59 22.5 N

Let’s encode the labels as \(1\) and \(0\) instead of E and N. Let’s do this below:

# Extract data for classification in numpy array
# Features
x = data['Height'].values
# Labels (must be integer)
label_coding = {'E': 1, 'N': 0}
y = np.array(
    [
        label_coding[r]
        for r in data['Result']
    ]
)
data['y'] = y
data
Hide code cell output
Height Result y
0 40.5 E 1
1 40.5 E 1
2 40.5 E 1
3 40.5 E 1
4 40.5 E 1
5 40.5 E 1
6 40.5 E 1
7 40.5 E 1
8 40.5 E 1
9 40.5 E 1
10 36.0 E 1
11 36.0 N 0
12 36.0 E 1
13 36.0 E 1
14 36.0 E 1
15 36.0 E 1
16 36.0 N 0
17 36.0 E 1
18 36.0 E 1
19 36.0 E 1
20 32.0 E 1
21 32.0 E 1
22 32.0 N 0
23 32.0 E 1
24 32.0 E 1
25 32.0 E 1
26 32.0 N 0
27 32.0 E 1
28 32.0 N 0
29 32.0 E 1
30 28.5 N 0
31 28.5 E 1
32 28.5 N 0
33 28.5 N 0
34 28.5 E 1
35 28.5 N 0
36 28.5 N 0
37 28.5 N 0
38 28.5 E 1
39 28.5 N 0
40 25.5 N 0
41 25.5 N 0
42 25.5 N 0
43 25.5 N 0
44 25.5 N 0
45 25.5 N 0
46 25.5 E 1
47 25.5 N 0
48 25.5 N 0
49 25.5 N 0
50 22.5 N 0
51 22.5 N 0
52 22.5 N 0
53 22.5 N 0
54 22.5 N 0
55 22.5 N 0
56 22.5 N 0
57 22.5 N 0
58 22.5 N 0
59 22.5 N 0

Let’s visualize the data. Notice that lots of observations fall on top of each other.

fig, ax = plt.subplots()
ax.plot(x, y, 'o')
ax.set_xlabel('$x$ (Height in cm)')
ax.set_ylabel('Result ($0=N; 1=E$)')
sns.despine(trim=True);
../_images/43efbb28eb411647980f8027b3e3d2e920854d3175db47f4b97068cabd0f87cc.svg

Let’s train a logistic regression model with just a linear feature using scikit-learn:

from sklearn.linear_model import LogisticRegression

# The design matrix
X = np.hstack(
    [
        np.ones((x.shape[0], 1)),
        x[:, None]]
)

# Train the model (penalty = 'none' means that we do not add a prior on the weights)
# we are effectively just maximizing the likelihood of the data
model = LogisticRegression(
    penalty=None,
    fit_intercept=False
).fit(X, y);

Here is how you can get the trained weights of the model:

model.coef_
array([[-12.688,   0.411]])

And here is how you can make predictions at some arbitrary heights:

x_predict = np.array([10.0, 20.0, 30.0, 40.0, 50.0])
X_predict = np.hstack(
    [
        np.ones((x_predict.shape[0], 1)),
        x_predict[:, None]
    ]
)
predictions = model.predict_proba(X_predict)
predictions
array([[9.99810986e-01, 1.89014081e-04],
       [9.88560090e-01, 1.14399105e-02],
       [5.85351787e-01, 4.14648213e-01],
       [2.25419752e-02, 9.77458025e-01],
       [3.76605766e-04, 9.99623394e-01]])

Note that the model gave us back the probability of each class. If you wanted, you could ask for the class of maximum probability for each prediction input:

model.predict(X_predict)
array([0, 0, 0, 1, 1])

To visualize the predictions of the model as a function of the height, we can do this:

fig, ax = plt.subplots()
xx = np.linspace(20.0, 45.0, 100)
XX = np.hstack([np.ones((xx.shape[0], 1)), xx[:, None]])
predictions_xx = model.predict_proba(XX)
ax.plot(
    xx,
    predictions_xx[:, 0],
    label='Probability of N'
)
ax.plot(
    xx,
    predictions_xx[:, 1],
    label='Probability of E'
)
ax.set_xlabel('$x$ (cm)')
ax.set_ylabel('Probability')
plt.legend(loc='best', frameon=False)
sns.despine(trim=True);
../_images/4ca08b6a49892708bfe94cb855a6239f08dad151118d7ba58b8e2a6843349780.svg

Questions#

  • What is the probability of an explosion when the height becomes very small?

  • What is the probability of an explosion when the height becomes very large?

  • At what height is it particularly difficult to predict what will happen?