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

# In[3]:


import numpy as np
import pickle, gzip


# In[4]:


f = gzip.open('mnist.pkl.gz', 'rb')
train_set, valid_set, test_set = pickle.load(f, encoding='latin1')
f.close()


# In[5]:


def isNumber(x, y):
    if(x == y):
        return 1
    else:
        return 0

def func(x):
    if(x >= 0):
        return 1.0
    else:
        return 0.0


# In[9]:


train_set[1][:10]


# In[7]:


def perceptron(train_set, target_set, number):
    image_set = train_set
    target_set = target_set
    target_set = np.vectorize(isNumber)(target_set, number)

    weight = np.zeros([785, ], dtype = float)
    image_set = np.insert(image_set, 0, 1, axis = 1)
    rate = 0.1


    output_func = np.vectorize(func)

    for i in range(len(image_set)):
        y = np.dot(image_set, weight)
        predict = output_func(y)

        weight +=  rate * image_set[i] * (target_set[i] - predict[i])
        if(i == 100):
            break
    return weight


# In[17]:


w0 = perceptron(train_set[0], train_set[1], 0)
w1 = perceptron(train_set[0], train_set[1], 1)
w2 = perceptron(train_set[0], train_set[1], 2)
w3 = perceptron(train_set[0], train_set[1], 3)
w4 = perceptron(train_set[0], train_set[1], 4)
w5 = perceptron(train_set[0], train_set[1], 5)
w6 = perceptron(train_set[0], train_set[1], 6)
w7 = perceptron(train_set[0], train_set[1], 7)
w8 = perceptron(train_set[0], train_set[1], 8)
w9 = perceptron(train_set[0], train_set[1], 9)


# In[16]:


t = np.insert(train_set[0], 0, 1, axis = 1)
func(np.dot(t[0], w0))


# In[27]:


def predictNumber(image):
    weight = [w0, w1, w2, w3, w4, w5, w6, w7, w8, w9]
    pred = []
    for w in weight:
        pred.append(func(np.dot(image, w)))
        #print(func(np.dot(image, w)))
    for j, i in enumerate(pred):
        if(i == 1):
            return j


# In[32]:


predictNumber(t[8])


# In[26]:


train_set[1][:20]


# In[ ]: