#!/usr/bin/env python
# coding: utf-8

# # MNIST digits: Softmax Regression 

# In[2]:


import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import fetch_openml # MNIST data
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from sklearn.metrics import accuracy_score


# In[3]:


# load MNIST data from https://www.openml.org/d/554
X, y = fetch_openml('mnist_784', version=1, return_X_y=True)
X = X.reshape((X.shape[0], -1))


# In[4]:


# generate training and test datasets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)


# In[5]:


plt.subplots(1, 6, figsize=(15,5))
for i in range(6):
    index = 9000 + i # image number
    pixels = np.array(X_train[index], dtype='uint8')
    pixels = pixels.reshape((28, 28))
    plt.subplot(1, 6, i+1)
    plt.title('Label is {label}'.format(label=y_train[index]))
    plt.imshow(pixels, cmap='gray')
    plt.xticks(())
    plt.yticks(())


# In[6]:


# apply logistic regressor with 'sag' solver, C is the inverse regularization strength
clf = LogisticRegression(multi_class='multinomial',
                         penalty='none', solver='sag', tol=0.1)


# In[7]:


# fit data
clf.fit(X_train, y_train)


# In[8]:


#Test the model
predictions = clf.predict(X_test)

#Precision, recall, f1-score
print(classification_report(y_test, predictions))
print(accuracy_score(y_test, predictions))


# In[9]:


import png
filename = "mnist_my_digit_3.png"
image = np.zeros((1, 28, 28, 1), dtype=np.uint8)
pngdata = png.Reader(open(filename, 'rb')).asDirect()
for i_row, row in enumerate(pngdata[2]):
    image[0, i_row, :, 0] = row

plt.imshow(np.squeeze(image), cmap="gray")
plt.xticks(())
plt.yticks(())
plt.show()

# one digit, -1: unspecified number determined by numpy
my_X = image.reshape((1,-1))


# In[11]:


probabilities = clf.predict_proba(my_X)
prediction = np.argmax(probabilities)
print(probabilities)
print(f"prediction = {prediction}")