Comparison of Neural Networks EE Code

import os
def loadRes(): #Loading data from txt file
    if(os.stat("results.txt").st_size == 0):
        return False
    f = open("results.txt")
    rawdat = f.read()

    f.close()
    dat = rawdat.split("\n")
    dat.pop()
    for i in range(0,len(dat)):
        dat[i] = dat[i].split("|")
        #print(dat[i])
        dat[i][2] = dat[i][2].split("~")
    return dat

def getNets(res):
    #print(res)
    if(res == False):
        return []
    results = []
    for i in res:
        results.append(i[2])
    return results

#Stackoverflow example of how to convert bases easily https://stackoverflow.com/questions/28824874/pythonic-way-to-do-base-conversion
def digit_to_char(digit):
    if digit < 10:
        return str(digit)
    return chr(ord('a') + digit - 10)

def str_base(number, base):
    while number > 0:
        number, digit = divmod(number, base)
        yield digit_to_char(digit)

def permuter(layers,nNodes):#Generate possible neural network combinations
    store = []
    check =  []
    for i in range(0,nNodes**layers):
        val = (''.join(list(str_base(i,nNodes))))
        if(val == ''):
            val = "0"
        val = format(int(val[::-1]), '0'+str(layers))
        if((val.replace("0","") in check) == False):
            store.append(list(val))
            check.append(val.replace("0",""))
    return store

import tensorflow as tf
from tensorflow.keras import backend as K
mnist = tf.keras.datasets.mnist #28x28

(trainx, trainy), (testx, testy) = mnist.load_data()

trainx = tf.keras.utils.normalize(trainx,axis=1)
testx = tf.keras.utils.normalize(testx,axis=1)
tf_config = tf.ConfigProto(allow_soft_placement=False)
tf_config.gpu_options.allow_growth = True
s = tf.Session(config=tf_config)
K.set_session(s)

def addLayers(layerData,neuronMultiplyer,model,base):#Add layers based on structure input
    model.add(tf.keras.layers.Flatten())
    for layer in layerData:
        layer = int(layer,base)
        if(layer!= 0):
            model.add(tf.keras.layers.Dense((layer)*neuronMultiplyer,activation=tf.nn.relu,kernel_initializer=tf.keras.initializers.glorot_uniform(1234),bias_initializer=tf.keras.initializers.glorot_uniform(1234)))

    return model

import os,datetime,timeit,time

def train(layerData,neuronMultiplyer,base):#Train and create neural network based on structure input
    f = open("results.txt", "a")
    os.system('cls')
    print(layerData)
    tf.keras.backend.clear_session() #Reset and clear session
    model = tf.keras.models.Sequential()
    model = addLayers(layerData,neuronMultiplyer,model,base)
    model.add(tf.keras.layers.Dense(10,activation=tf.nn.softmax,kernel_initializer=tf.keras.initializers.glorot_uniform(1234),bias_initializer=tf.keras.initializers.glorot_uniform(1234)))
    model.compile(optimizer='adam',
                loss='sparse_categorical_crossentropy',
                 metrics=['accuracy'])

    model.fit(trainx,trainy,epochs=3,batch_size=256,use_multiprocessing=True, workers=10)
    start = time.perf_counter()
    valLoss,valAcc = model.evaluate(testx,testy)
    elapsed = time.perf_counter() - start #10000 samples takes this long.
    f.write(str(valLoss)+"|"+str(valAcc)+"|"+'~'.join(layerData)+"|"+str(elapsed)+"\n")
    return valLoss,valAcc,layerData

def evaluate(layers,maxNeurons,inc):
    data = permuter(layers,int(maxNeurons//inc +1))
    fileDat = getNets(loadRes())
    for structure in data:
        if((structure in fileDat)==False):
            train(structure,inc,int(maxNeurons//inc +1))

evaluate(5, 100, 20)
input("Process is done, press enter to close.")