Tag: python

Tested Bard vs ChatGPT to write a Python script for MNIST digit prediction!

I had been using ChatGPT and then Bard for ‘inspiration’ when doing some of our Python scripting on newer stuff that I have been learning. It’s like having an over-caffeinated coding pal next to you to hammer out any test code you fancy.

I had some “help” from ChatGPT when learning to connect and read sensor data from a Silicon Labs XG24 Bluetooth/BLE development board and the baseline idea for our team’s further development is the bag.

OK, this experiment would be interesting .. I like to see how “correct” the first response from Bard and ChatGPT to write a Python script basically to grab an unknown sized “written digit” image from my drive and predict it against the MNIST database using a tensorflow lite model file that I already prepared. It will also need to calculate elapsed time.

Test digit image vs. the MNIST digit database images

My ‘one-shot’ prompt is ..

you are an experienced python machine learning developer. you are to write python 3.9 script to predict and print out the the digit of a hand-written image from a file “web_four.png”and classify the image against the “mnist.tflite” model in the same folder. you need to calculate the elapsed time for the inference. before inference, you will need to reshape the image file to 28×28 and 32-bit floating point grey-scale

And I unleashed the machines …

A second or two later …

ChatGPT started to ‘write’ line by line while Bard paused a second and off-loaded the whole script onto the screen. I reckon “they” both took about the same time to complete the job. Also some advise to make sure of file location and type etc from each of them.

The script from Bard …

import tensorflow as tf
import time
# Load the MNIST model
model = tf.lite.load_model("mnist.tflite")
# Read the hand-written image file
image = tf.io.read_file("web_four.png")
# Convert the image to a 28x28 32-bit floating point grey-scale tensor
image = tf.image.decode_png(image, channels=1)
image = tf.image.resize(image, (28, 28))
image = tf.cast(image, tf.float32)
# Start the timer
start = time.time()
# Classify the image
prediction = model.predict(image)
# Stop the timer
end = time.time()
# Print the predicted digit and the elapsed time
print("Predicted digit:", prediction[0])
print("Elapsed time:", end - start)

.. and the code from ChatGPT

import numpy as np
import tensorflow as tf
from PIL import Image
import time
# Load the model
interpreter = tf.lite.Interpreter(model_path='mnist.tflite')
interpreter.allocate_tensors()
# Load and preprocess the image
image_path = 'web_four.png'
image = Image.open(image_path).convert('L')
image = image.resize((28, 28))
image = np.array(image).astype('float32') / 255.0
image = np.expand_dims(image, axis=0)
image = np.expand_dims(image, axis=-1)
# Get input and output tensors
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
# Set the input tensor
interpreter.set_tensor(input_details[0]['index'], image)
# Run the inference
start_time = time.time()
interpreter.invoke()
end_time = time.time()
elapsed_time = end_time - start_time
# Get the output tensor
output = interpreter.get_tensor(output_details[0]['index'])
# Get the predicted digit
predicted_digit = np.argmax(output)
# Print the results
print(f'Predicted Digit: {predicted_digit}')
print(f'Inference Time: {elapsed_time} seconds')

Now judgement time for this experiment .. it’s brutal I guess, one-shot only and loser goes home hhh .. let’s run Bard’s response first 

stanley@stanLG:~/tf$ python3 bard_mnist.py
2023-05-13 09:45:56.825094: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
Traceback (most recent call last):
  File "/home/stanley/tf/bard_mnist.py", line 5, in <module>
    model = tf.lite.load_model("mnist.tflite")
AttributeError: module 'tensorflow._api.v2.lite' has no attribute 'load_model'

and ChatGPT’s attempt

stanley@stanLG:~/tf$ python3 chatgpt_mnist.py
2023-05-13 09:46:38.712179: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
Predicted Digit: 6
Inference Time: 0.00041866302490234375 seconds

Oh! Bard somehow have made a mistake with a module attribute and was not able to delivery a working solution at its first attempt while ChatGPT was spot-on with a working script delivering the required process (even though the prediction was off due to my shabby training …)

So, for now it’s 0-1 with “old man” ChatGPT from OpenAI ahead this round.

Before we go, I gave Bard a chance to redeem “himself” …


.. let’s try the fix from Bard

Traceback (most recent call last):
  File "/home/stanley/tf/bard_mnist.py", line 5, in <module>
    model = tf.io.parse_and_load_tflite_model("mnist.tflite")
AttributeError: module 'tensorflow._api.v2.io' has no attribute 'parse_and_load_tflite_model'

Oh no!!! … not sure where Bard’s been picking his bad habit from this morning but he still stuffed it up. Initially was my fault for not explaining that we are using Tensorflow 2.X (which should be the case for anyone coding today …. ) but sorry “No Pass” Bard this round.

Related posts

Latest posts