cihat645

//
//  main.swift
//  LinearRegression
//
//  Created by Thomas Ciha on 5/20/17.
//  Copyright © 2017 Thomas Ciha. All rights reserved.
//

import Foundation

cihat645

Offer(float p, int q, milliseconds m, Term t, OrdType ot, bool b, int oID){
    Price = p;
    Qty = q;
    OrderID = oID;
    timestamp = m;
    OfferTerm = t;
    OrderType = ot;
    my_offer = b;                                                                                       
}

cihat645

order_info(float f, int q, Status s, int oID, bool flag){
  avg_price = f;
  shares_filled = q;
  offer_status = s;
  order_ID = oID;
  incomplete_market_order = flag;
 }

cihat645

def create_five_nns(input_size, hidden_size, act = None):
    """
    Creates 5 neural networks to be used as a baseline in determining the influence model depth & width has on performance.
    :param input_size: input layer size
    :param hidden_size: list of hidden layer sizes
    :param act: activation function to use for each layer
    :return: list of model_info hash tables
    """
    act = ['relu'] if not act else [act]                             # default activation = 'relu'
    nns = []                                                         # list of model info hash tables

cihat645

def quick_nn_test(model_info, data_dict, save_path):
    model = build_nn(model_info)                                    # use model info to build and compile a nn
    stop = EarlyStopping(patience=5, monitor='acc', verbose=1)      # maintain a max accuracy for a sliding window of 5 epochs. If we cannot breach max accuracy after 15 epochs, cut model off and move on.
    tensorboard_path =save_path + model_info['Name']                # create path for tensorboard callback
    tensorboard = TensorBoard(log_dir=tensorboard_path, histogram_freq=0, write_graph=True, write_images=True)              # create tensorboard callback
    save_model = ModelCheckpoint(filepath= save_path + model_info['Name'] + '\\' + model_info['Name'] + '_saved_' + '.h5')  # save model after every epoch

    model.fit(data_dict['Training data'], data_dict['Training labels'], epochs=150,               # fit model
              batch_size=model_info['Batch size'], callbacks=[save_model, stop, tensorboard])     # evaluate train accuracy
    

cihat645

"""This section of code allows us to create and test many neural networks and save the results of a quick 
test into a CSV file. Once that CSV file has been created, we will continue to add results onto the existing 
file."""

rapid_testing_path = 'YOUR PATH HERE'  
data_path = 'YOUR DATA PATH'

try:                                                                        # try to load existing csv
    rapid_mlp_results = pd.read_csv(rapid_testing_path + 'Results.csv')
    index = rapid_mlp_results.shape[1]

cihat645

def generate_random_model():
    optimization_methods = ['adagrad', 'rmsprop', 'adadelta', 'adam', 'adamax', 'nadam']      # possible optimization methods
    activation_functions = ['sigmoid', 'relu', 'tanh']          # possible activation functions
    batch_sizes = [16, 32, 64, 128, 256, 512]                   # possible batch sizes
    range_hidden_units = range(5, 250)                          # range of possible hidden units
    model_info = {}                                             # create hash table
    same_units = np.random.choice([0, 1], p=[1/5, 4/5])         # dictates whether all hidden layers will have the same number of units
    same_act_fun = np.random.choice([0, 1], p=[1/10, 9/10])     # will each hidden layer have the same activation function?
    really_deep = np.random.rand()
    range_layers = range(1, 10) if really_deep < 0.8 else range(6, 20)          # 80% of time constrain number of hidden layers between 1 - 10, 20% of time permit really deep architectures

cihat645

def build_nn(model_info):
    """
    This function builds and compiles a NN given a hash table of the model's parameters.
    :param model_info:
    :return:
    """

    try:
        if model_info["Regularization"] == "l2":                                # if we're using L2 regularization