Introduction to Tensorflow¶

  • Simple linear regression with Tensorflow.
    • Fit a linear regression line
    • Classify images
  • Visualizing Training Progress with Tensorboard.
  • Re-using models - Pretrained models from Tensorflow Model Hub
    • Fine tune an image model to classify images
In [88]:
Copied 
# ! pip install -q seaborn tensorflow_hub
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
import requests

# Make NumPy printouts easier to read.
np.set_printoptions(precision=3, suppress=True)

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

print(tf.__version__)

2.8.0

Utility Functions¶

Implement a few utility functions which we will use across the notebook

In [89]:
Copied 
# plot model training history
def plot_history(history):
    
    metrics =  [x for x in history.history.keys() if "val" not in x]
    # metrics = list(history.history.keys())
    ncols = 2
    nrows = int(np.ceil(len(metrics)/ncols))  
    fig, ax = plt.subplots(nrows=nrows, ncols=ncols, figsize=(18,nrows*6) )
    fig.suptitle(f"Model Training History" )
    i = 0 
    if ax.ndim == 1:
        ax = ax.reshape(ax.shape[0],1) 
    for row in ax:
        for col in row:
            col.plot(history.history[metrics[i]])
            col.plot(history.history["val_"+metrics[i]])
            col.set_xlabel("Epoch")
            col.set_ylabel(f"{metrics[i]}")
            col.legend(['train', 'val'], loc='upper right')
            col.title.set_text(f"{metrics[i]}")
            i += 1
    plt.tight_layout()
    plt.rcParams.update({'font.size': 18})  
    
    plt.show()

# download data file
def download_file(url): 
  download_path = f'/tmp/{url.split("/")[-1]}'
  results = requests.get(url)
  with open(download_path, 'wb') as f:
      f.write(results.content)
  return download_path

Linear Regression with Tensorflow¶

Based on Tensorflow tutorial here.

  • Download the MPG Dataset
  • Build a neural network model to predict Miles Per Gallon for a car, given its properties.
In [90]:
Copied 
# download CSV
url = 'http://archive.ics.uci.edu/ml/machine-learning-databases/auto-mpg/auto-mpg.data'
column_names = ['MPG', 'Cylinders', 'Displacement', 'Horsepower', 'Weight',
                'Acceleration', 'Model Year', 'Origin']

# Read csv file using pandas
raw_dataset = pd.read_csv(url, names=column_names,
                          na_values='?', comment='\t',
                          sep=' ', skipinitialspace=True)
dataset = raw_dataset.copy()
dataset = dataset.dropna()

# Preprocessing to replace car origin field one hot encoded variables
dataset['Origin'] = dataset['Origin'].map({1: 'USA', 2: 'Europe', 3: 'Japan'})
dataset = pd.get_dummies(dataset, columns=['Origin'], prefix='', prefix_sep='')

# Split dataset into train and test split
train_dataset = dataset.sample(frac=0.8, random_state=0)
test_dataset = dataset.drop(train_dataset.index)

train_features = train_dataset.copy()
test_features = test_dataset.copy()

train_labels = train_features.pop('MPG')
test_labels = test_features.pop('MPG')


# Build a feature normalization layer. Fit it on train data only
normalizer = tf.keras.layers.Normalization(axis=-1)
normalizer.adapt(np.array(train_features))
In [91]:
Copied 
dataset
Out[91]:
MPG Cylinders Displacement Horsepower Weight Acceleration Model Year Europe Japan USA
0 18.0 8 307.0 130.0 3504.0 12.0 70 0 0 1
1 15.0 8 350.0 165.0 3693.0 11.5 70 0 0 1
2 18.0 8 318.0 150.0 3436.0 11.0 70 0 0 1
3 16.0 8 304.0 150.0 3433.0 12.0 70 0 0 1
4 17.0 8 302.0 140.0 3449.0 10.5 70 0 0 1
... ... ... ... ... ... ... ... ... ... ...
393 27.0 4 140.0 86.0 2790.0 15.6 82 0 0 1
394 44.0 4 97.0 52.0 2130.0 24.6 82 1 0 0
395 32.0 4 135.0 84.0 2295.0 11.6 82 0 0 1
396 28.0 4 120.0 79.0 2625.0 18.6 82 0 0 1
397 31.0 4 119.0 82.0 2720.0 19.4 82 0 0 1

392 rows × 10 columns

In [92]:
Copied 
linear_model = tf.keras.Sequential([
    normalizer,
    layers.Dense(units=1)
])
linear_model.compile(
    optimizer=tf.optimizers.Adam(learning_rate=0.06),
    loss='mape', metrics=["mape"])

history = linear_model.fit(
    train_features,
    train_labels,
    verbose=0,
    epochs=130,  
    validation_split = 0.2)

plot_history(history)

Get Predictions on Test Set¶

Get predictions on the test set, compare with original labels

In [93]:
Copied 
predictions = linear_model.predict(test_features) 
test_preds= pd.DataFrame({"Original MPG": list(test_labels), "Predicted MPG": list(predictions.reshape(-1))})
test_preds["Percentage Difference"] = round( ( abs(test_preds["Original MPG"] - test_preds[ "Predicted MPG"]) / test_preds["Original MPG"])*100 ,2)
test_preds
Out[93]:
Original MPG Predicted MPG Percentage Difference
0 15.0 14.344485 4.37
1 10.0 9.327795 6.72
2 9.0 8.066470 10.37
3 25.0 27.019777 8.08
4 19.0 20.926760 10.14
... ... ... ...
73 34.0 27.251884 19.85
74 36.0 31.008327 13.87
75 34.0 32.229446 5.21
76 32.0 33.706100 5.33
77 28.0 26.550909 5.18

78 rows × 3 columns

In [94]:
Copied 
plt.figure(figsize=(18,6))
plt.hist(test_preds["Percentage Difference"], bins=40);
plt.title("Histogram of Mean Absolute Error");