import pandas as pd
import numpy as np
from sklearn.naive_bayes import MultinomialNB
from sklearn.svm import SVC
from google.colab import files
import matplotlib.pyplot as plt
import csv
import seaborn as sns

files.upload()

files.upload()

# READ THROUGH EACH LINE IN THE TRAINING CSV

train_image_list = []
train_label_list = []
with open("fashion-mnist_train.csv", "r") as train_file:
  # CREATE THE READER THAT READS EACH LINE
  reader = csv.reader(train_file, delimiter = ",")
  # THIS IS HOW YOU SKIP OVER A LINE
  # THE LINE BEING SKIPPED IS THE HEADER (COLUMN NAMES)
  next(reader)
  for row in reader:
    # THE LABEL IS THE FIRST ELEMENT IN EACH ROW
    # APPEND EACH LABEL TO ITS OWN ROW
    label = row[0]
    train_label_list.append(label)
    # DELETE THE LABEL FROM EACH LIST
    row.pop(0)
    # WHATS LEFT IN THE LIST IS JUST THE PIXELS FOR THE IMAGES (NO LABELS)
    train_image_list.append(row)

# READ THROUGH EACH LINE IN THE TEST CSV

test_image_list = []
test_label_list = []
with open("fashion-mnist_test.csv", "r") as test_file:
  # CREATE THE READER THAT READS EACH LINE
  reader = csv.reader(test_file, delimiter = ",")
  # THIS IS HOW YOU SKIP OVER A LINE
  # THE LINE BEING SKIPPED IS THE HEADER (COLUMN NAMES)
  next(reader)
  for row in reader:
    # THE LABEL IS THE FIRST ELEMENT IN EACH ROW
    # APPEND EACH LABEL TO ITS OWN ROW
    label = row[0]
    test_label_list.append(label)
    # DELETE THE LABEL FROM EACH LIST
    row.pop(0)
    # WHATS LEFT IN THE LIST IS JUST THE PIXELS FOR THE IMAGES (NO LABELS)
    test_image_list.append(row)

# CONVERT THE ELEMENTS IN THE LIST FROM STR --> INT

train_labels = []
for elem in train_label_list:
  elem = int(elem)
  train_labels.append(elem)

# CONVERT THE ELEMENTS IN EACH LIST FROM STR --> INT

train_images = []
for item in train_image_list:
  for i in range(len(item)):
    item[i] = int(item[i])

train_data_with_label = pd.read_csv("fashion-mnist_train.csv")
train_data_with_label.head()

test_data_with_label = pd.read_csv("fashion-mnist_test.csv")
test_data_with_label.head()

# SHOW AN EXAMPLE IMAGE OF EACH TYPE OF CLOTHING IN THE DATASET



# CREATE 10 SUBPLOTS, 2 ROWS AND 5 COLUMNS LONG WITH THE SAME X AND Y AXIS
fig, ax = plt.subplots(nrows = 2, ncols = 5, sharex = True, sharey = True)
ax = ax.flatten()
for i in range(0,10):
  # FIND THE INDEX VALUE OF EVERY DIFFERENT LABEL (0-9)
    # only finds the first occurance in this case
  # THIS INDEX MATCHES UP WITH A PICTURE IN THE IMAGES LIST
  image = train_image_list[train_labels.index(i)]

  # CONVERT THE LIST OF PIXELS INTO A NUMPY ARRAY
    # a list cannot be reshaped, must be an array
    # go from (1, 754) to (28,28)
  # RESHAPE THE ARRAY TO THE SIZE OF THE PICTURE (28,28) IN THIS CASE
  image = np.asarray(image).reshape(28,28)

  # EACH DIFFERENT AXIS WILL BE A DIFFERENT PIECE OF CLOTHING
  ax[i].imshow(image, cmap = "Greys", interpolation = "nearest")

# GET RID OF THE X AND Y TICKS
ax[0].set_xticks([])
ax[0].set_yticks([])
plt.tight_layout()
plt.show()

# CREATE A PLOT WITH THE COUNTS OF EACH LABEL
counts = sns.countplot(x = "label", data = train_data_with_label)

# CHANGE THE TITLE OF THE CHART
counts.set_title("Counts for each label")

# SET THE POSITION OF THE TICKS
  # should be the same length of the number of bars
counts.set_xticks([0,1,2,3,4,5,6,7,8,9])

# CHANGE THE LABELS OF THE TICKS
# ROTATE THE LABELS SO THEY CAN BE READ AND DO NOT OVERLAP EACH OTHER
counts.set_xticklabels(["T-Shirt", "Trousers", "Pullover", "Dress", "Coat", "Sandal", "Shirt", "Sneaker", "Bag", "Ankle Boot"], 
                       rotation = 45)
                       

training_labels = train_data_with_label["label"]
training_labels.reset_index(inplace = True, drop = True)
training_data = train_data_with_label.drop("label", axis = 1, inplace = False)
training_data.reset_index(inplace = True, drop = True)

print("Train label shape: ", training_labels.shape)
print("Train data shape: ", training_data.shape)

testing_labels = test_data_with_label["label"]
testing_labels.reset_index(inplace = True, drop = True)
testing_data = test_data_with_label.drop("label", axis = 1, inplace = False)
testing_labels.reset_index(inplace = True, drop = True)

print("Test label shape: ", testing_labels.shape)
print("Test data shape: ", testing_data.shape)

from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score

nb = MultinomialNB()
nb_model = nb.fit(training_data, training_labels)
nb_pred = nb_model.predict(testing_data)
nb_confusion = confusion_matrix(testing_labels, nb_pred)
nb_accuracy = accuracy_score(testing_labels, nb_pred)
print(nb_confusion)
print("\nNaive Bayes Accuracy: %", nb_accuracy * 100)