Python

Using the code below

$1.$ adjust parameters to improve accuracy and a discussion on the activities you performed and results.

$2.$ Generate a Tensorboard and produce graphics that show various improvements that you were able to gain. Loss and Accuracy for train and validation for example.

#Load the libraries $-$ Load them all here $-$ not sprinkled throughout the notebook.

import pandas as pd

import numpy as np

from sklearn.model$\_$selection import train$\_$test$\_$split

from sklearn.preprocessing import StandardScaler

import matplotlib.pyplot as plt

import seaborn as sns

from sklearn.pipeline import Pipeline

from sklearn.decomposition import PCA

from sklearn.preprocessing import OneHotEncoder

from sklearn.model$\_$selection import GridSearchCV

from sklearn.metrics import accuracy$\_$score, confusion$\_$matrix, classification$\_$report, recall$\_$score, precision$\_$score, f$1\_$score

import warnings

warnings.filterwarnings$(\text{'}$ignore$\text{'})$ #ignore warnings

$\%$load$\_$ext tensorboard

import tensorflow as tf

from tensorflow.keras.callbacks import EarlyStopping

from tensorflow.keras.optimizers import Adam, Nadam, RMSprop

mnist $=$ tf$.$keras.datasets.mnist

$($x$\_$train, y$\_$train$),($x$\_$test, y$\_$test$)=$ mnist.load$\_$data$()$

from google.colab import drive

drive.mount$(\text{'}/$content$/$drive$\text{'})$

df$\_$mnist$\_$train $=$ pd$.$read$\_$csv$(\text{'}/$content$/$drive$/$MyDrive$/$Colab Notebooks$/$data$/$mnist$\_$train.csv$\text{'})$

df$\_$mnist$\_$test $=$ pd$.$read$\_$csv$(\text{'}/$content$/$drive$/$MyDrive$/$Colab Notebooks$/$data$/$mnist$\_$test.csv$\text{'})$

Model Development

batch$\_$size $=32$

num$\_$classes $=10$

epochs $=10$

input$\_$shape $=(28,28,1)$

optimizer $=$ Adam$($learning$\_$rate$=0\mathrm{.}001)$

log$\_$dir $=$ "logs$/$fit$/"$ # Adjust the directory as needed

tensorboard$\_$callback $=$ tf$.$keras.callbacks.TensorBoard$($log$\_$dir$=$log$\_$dir, histogram$\_$freq$=1)$

model $=$ tf$.$keras.models.Sequential$([$

tf$.$keras.layers.Conv$2$D$(32,(5,5),$ padding$=$'same', activation$=$'relu', input$\_$shape$=$input$\_$shape$),$

tf$.$keras.layers.Conv$2$D$(32,(5,5),$ padding$=$'same', activation$=$'relu'$),$

tf$.$keras.layers.MaxPool$2$D$(),$

tf$.$keras.layers.Dropout$(0\mathrm{.}25),$

tf$.$keras.layers.Conv$2$D$(64,(3,3),$ padding$=$'same', activation$=$'relu'$),$

tf$.$keras.layers.Conv$2$D$(64,(3,3),$ padding$=$'same', activation$=$'relu'$),$

tf$.$keras.layers.MaxPool$2$D$($strides$=(2,2)),$

tf$.$keras.layers.Dropout$(0\mathrm{.}25),$

tf$.$keras.layers.Flatten$(),$

tf$.$keras.layers.Dense$(128,$ activation$=$'relu'$),$

tf$.$keras.layers.Dropout$(0\mathrm{.}5),$

tf$.$keras.layers.Dense$($num$\_$classes, activation$=$'softmax'$)$

$])$

model.compile$($optimizer$=$optimizer, loss$=$'categorical$\_$crossentropy', metrics$=[\text{'}$acc$\text{'}])$

# Define early stopping callback

early$\_$stopping $=$ EarlyStopping$($

monitor$=$'val$\_$loss', # Monitor validation loss

patience$=3,$ # Number of epochs with no improvement after which training will stop

min$\_$delta$=0\mathrm{.}001,$ # Minimum change to qualify as an improvement

mode$=$'min', # 'min' because we're looking for the minimum validation loss

restore$\_$best$\_$weights$=$True # Restore model weights from the epoch with the best validation loss

$)$

from sys import version

# Train the model with early stopping

history $=$ model.fit$($

train$\_$array, y$\_$train,

validation$\_$split$=0\mathrm{.}2,$

epochs$=$epochs,

batch$\_$size$=$batch$\_$size,

verbose $=1,$

callbacks$=[$early$\_$stopping, tensorboard$\_$callback$]$

$)$

fig, ax $=$ plt$.$subplots$(2,1)$

ax$[0].$plot$($history$.$history$[\text{'}$loss$\text{'}],$ color$=\text{'}$b$\text{'},$ label$=$"Training Loss"$)$

ax$[0].$plot$($history$.$history$[\text{'}$val$\_$loss'$],$ color$=\text{'}$r$\text{'},$ label$=$"Validation Loss",axes $=$ax$[0])$

legend $=$ ax$[0].$legend$($loc$=$'best', shadow$=$True$)$

ax$[1].$plot$($history$.$history$[\text{'}$acc$\text{'}],$ color$=\text{'}$b$\text{'},$ label$=$"Training Accuracy"$)$

ax$[1].$plot$($history$.$history$[\text{'}$val$\_$acc'$],$ color$=\text{'}$r$\text{'},$label$=$"Validation Accuracy"$)$

legend $=$ ax$[1].$legend$($loc$=$'best', shadow$=$True$)$

test$\_$loss, test$\_$acc $=$ model.evaluate$($test$\_$array, y$\_$test$)$

# Predict the values from the testing dataset

Y$\_$pred $=$ model.predict$($test$\_$array$)$

# Convert predictions classes to one hot vectors

Y$\_$pred$\_$classes $=$ np$.$argmax$($Y$\_$pred,axis $=1)$

# Convert testing observations to one hot vectors

Y$\_$true $=$ np$.$argmax$($y$\_$test,axis $=1)$

# compute the confusion matrix

confusion$\_$mtx $=$ tf$.$math.confusion$\_$matrix$($Y$\_$true, Y$\_$pred$\_$classes$)$

plt$.$figure$($figsize$=(10,8))$

sns$.$heatmap$($confusion$\_$mtx$,$ annot$=$True, fmt$=\text{'}$g$\text{'})$

accuracy $=$ accuracy$\_$score$($Y$\_$true, Y$\_$pred$\_$classes$)$

recall $=$ recall$\_$score$($Y$\_$true, Y$\_$pred$\_$classes, average$=$'weighted'$)$

precision $=$ precision$\_$score$($Y$\_$true, Y$\_$pred$\_$classes, average$=$'weighted'$)$

f$1=$ f$1\_$score$($Y$\_$true, Y$\_$pred$\_$classes, average$=$'weighted'$)$

print$($f$"$Accuracy: $\{$accuracy$\}")$

print$($f$"$Recall: $\{$recall$\}")$

print$($f$"$Precision: $\{$precision$\}")$

print$($f$"$F$1$ Score: $\{$f$1\}")$

$\%$tensorboard $--$logdir logs$/$fit

batch$\_$size $=32$

num$\_$classes $=10$

epochs $=10$

train$\_$array $=$ train$\_$array.reshape$(60000,28,28)$

optimizer $=$ Adam$($learning$\_$rate$=0\mathrm{.}001)$

log$\_$dir $=$ "logs$/$fit$/"$ # Adjust the directory as needed

tensorboard$\_$callback $=$ tf$.$keras.callbacks.TensorBoard$($log$\_$dir$=$log$\_$dir, histogram$\_$freq$=1)$

The code is continued on the image