Penguin activity

Penguin activity#

Background#

In this activity, we will install TensorFlow and use it to make a neural network for predicting Penguin species from the Palmer Penguin dataset. The data can be downloaded from here.

../_images/penguins.png

Artwork by @allison_horst

We talked about it in class but, as a reminder, this dataset contains attributes for 342 penguins collected from three islands in the Palmer Archipelago, Antarctica. More information about the Palmer Penguin dataset can be found here.

Our goal in this activity is to predict body mass using the other features in the dataset (i.e. culmen length, culmen depth, and flipper length). To do this, we will perform regression using a neural network.

Task 1 (0 points)#

We first need to install TensorFlow which, on MacOS at least, requires Python 3.9-3.11. The easiest way to do this is to open an Anaconda Prompt (miniconda3) (Windows) or Terminal (macOS) and create a new environment called tf:

conda create --name tf python=3.11

Activate the environment:

conda activate tf

  • Install tensorflow using pip:

pip install tensorflow

  • Install some other packages:

pip install notebook pandas matplotlib scikit-learn

  • Open a jupyter notebook

jupyter notebook

Tip

If you run this command from your course folder, your .ipynb assignment will automatically be saved there.

Task 2 (0 points)#

Import the Penguin dataset by running the following:

Note

You will need to change the path name to the csv file

# Import packages
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split

# Read training data
df = pd.read_csv(`data/penguins.csv')

# Define feature list
feature_list =  ['Culmen Length (mm)', 'Culmen Depth (mm)',
       'Flipper Length (mm)']
X = df[feature_list]
y = df['Body Mass (g)']

# Standarize features
x_scaler = StandardScaler()  
X_scaled = x_scaler.fit_transform(X)
X_scaled = pd.DataFrame(X_scaled, columns=X.columns)
X_scaled.index = X.index

y_scaler = StandardScaler()  
y_scaled = y_scaler.fit_transform(y.values.reshape(-1, 1))
y_scaled = pd.DataFrame(y_scaled.flatten(), index=y.index, columns=[y.name])

# Split into training, testing, and validation
X_train, X_test, y_train, y_test = train_test_split(
    X_scaled, y_scaled, test_size=0.2, random_state=42)

To expedite things, this code standardizes the data and defines training and testing datasets.

Task 3 (10 points)#

  • Build a feedforward neural network that predicts y_train from X_train using TensorFlow.

Note

The model should contain at least one hidden layer.

  • Evaluate the mean absolute error (MAE) of your model by running:

# Predict on test set
y_pred_scaled = model.predict(X_test).flatten()
y_pred = y_scaler.inverse_transform(y_pred_scaled.reshape(1, -1)).flatten()
y_actual = np.array(y_scaler.inverse_transform(y_test)).flatten()

# Report MAE
print('MAE = %.2f g' % (np.mean(np.abs(y_actual-y_pred))))

  • Plot the predictions of your model by running:

# Plot
fig, ax = plt.subplots(figsize=(10, 4))
ax.scatter(y_actual, y_pred, zorder=3)
ax.plot([2500, 6500], [2500, 6500], lw=2, zorder=3)
ax.set_ylabel('Predicted body mass (g)', fontsize=12)
ax.set_xlabel('Actual body mass (g)', fontsize=12)
ax.tick_params(axis='both', which='major', labelsize=12)
ax.grid(ls='dashed', lw=1, zorder=1)
ax.text(0.05, 0.95, f"MAE = {mae:.2f} g", transform=ax.transAxes, 
        fontsize=12, verticalalignment='top', 
        bbox=dict(boxstyle="round,pad=0.3", fc="white", 
              ec="black", alpha=0.7))

Task 3 (5 points)#

Tinker with your model to see if you can improve your MAE.

Important

Save your notebook locally in both .ipynb and .pdf formats but only submit the pdf to Canvas.