配列の治し方(？)がわかりません

エラーの内容を調べたところおそらく量が合っていないから計算をすることができないというところまではわかったのですが、どうすれば量を変えることができるのかがわかりません
reshapeを使えば変わるということまでは覚えているのですが、この場合どう書けばいいのかわかりませんでした。
教えてください。
お願いします。

# -*- coding: utf-8 -*-
import keras
from keras.datasets import mnist
import matplotlib 
from matplotlib import pyplot
import numpy as np
from sklearn import datasets
from sklearn.model_selection import cross_val_score as crv

def sigmoid(x):
    return 1 / (1 + np.exp(-x))

def softmax(x):
    expX = np.exp(x)
    return expX / np.sum(expX)

def cross_entropy_error(y,t):
    delta = 1e-7
    batch_size = y.shape[0]
    idx= np.arange(batch_size)
    return -np.sum(np.log(y[idx,t]+delta)) / batch_size

def gradient(f,x):
    if x.ndim == 1:
        return gradient_sub(f,x)
    else:
        grad = np.zeros_like(x)
        for index, xx in enumerate(x):
            grad[index] = gradient_sub(f,xx)
    return grad

def gradient_sub(f,x):
    h = 1e-4
    grad = np.zeros_like(x)
    for i in range(x.size):
        val = x[i]
        x[i] = val + h
        fx1 = f(x)
        x[i] = val - h
        fx2 = f(x)
        grad[i] = (fx1 - fx2) / (2*h)
        x[i] = val
    return grad


class SampleNetwork:
    def __init__(self):
        input_size = 64
        hidden_size = 50
        output_size = 10
        self.params = {}
        self.params["w0"] = 0.01 * np.random.randn(input_size,hidden_size)
        self.params["w1"] = 0.01 * np.random.randn(hidden_size,output_size)
        self.params["b0"] = np.zeros(hidden_size)
        self.params["b1"] = np.zeros(output_size)
        self.learning_rate = 0.1

    def predict(self,x):
        a0 = np.dot(x,self.params["w0"]) + self.params["b0"]
        z0 = sigmoid(a0)
        a1 = np.dot(z0,self.params["w1"]) + self.params["b1"]
        y = softmax(a1)
        return y

    def update_params(self,x,t):
        loss_W = lambda W: self.loss(x,t)
        for key in self.params.keys():
            grad = gradient(loss_W, self.params[key])
            self.params[key] -= self.learning_rate*grad


    def loss(self,x,t):
        y = self.predict(x)
        return cross_entropy_error(y,t)

    def accurary(self,x,t):
        y = self.predict(x)
        y = np.argmax(y,axis=1)
        acc = np.sum(y==t) / float(x.shape[0])
        return acc

digits = datasets.load_digits()
X = digits.data
T = digits.target

(train_X, train_T), (test_X, test_T) = mnist.load_data()

network = SampleNetwork()

batch_size = 100
iter_num = 300
train_size = train_X.shape[0]
erpoch_size = max(train_size//batch_size,1)
loss_list = []
train_accurary_list = []
test_accurary_list = []

for index in range(iter_num):
    batch_choice = np.random.choice(train_size,batch_size)
    x_batch = train_X[batch_choice]
    t_batch = train_T[batch_choice]
    network.update_params(x_batch,t_batch)
    loss = network.loss(x_batch,t_batch)
    loss_list.append(loss)
    if (index % erpoch_size == 0):
        train_accurary = network.accurary(train_X,train_T)
        test_accurary = network.accurary(test_X,test_T)
        train_accurary_list.append(train_accurary)
        test_accurary_list.append(test_accurary)

pyplot.figure(figsize=(10,7))
pyplot.subplot(2,2,1)
pyplot.plot(np.arange(len(loss_list)),loss_list)
pyplot.xlabel("iteration")
pyplot.title("Cross Entropy Error")
pyplot.subplot(2,2,2)
pyplot.plot(np.arange(0,len(train_accurary_list),1),train_accurary_list,"b")
pyplot.plot(np.arange(0,len(test_accurary_list),1),test_accurary_list,"ro")
pyplot.xlabel("iteration(epoch)")
pyplot.title("Accuary")
pyplot.legend(("train","test"),loc = "lowrer right")
pyplot.tight_layout()

エラーが
Using TensorFlow backend.
Traceback (most recent call last):
File "number.py", line 104, in <module>
network.update_params(x_batch,t_batch)
File "number.py", line 70, in update_params
grad = gradient(loss_W, self.params[key])
File "number.py", line 44, in gradient
grad[index] = gradient_sub(f,xx)
File "number.py", line 31, in gradient_sub
fx1 = f(x)
File "number.py", line 68, in <lambda>
loss_W = lambda W: self.loss(x,t)
File "number.py", line 75, in loss
y = self.predict(x)
File "number.py", line 61, in predict
a0 = np.dot(x,self.params["w0"]) + self.params["b0"]
ValueError: shapes (100,28,28) and (64,50) not aligned: 28 (dim 2) != 64 (dim 0)
と出ます