val_data['return'] = df.loc[val_index, 'return'] mu = np.mean(est_data['return']) est_data['return_dm'] = est_data['return'] - mu val_data['return_dm'] = val_data['return'] - mu est_data['return_dm2'] = est_data['return_dm']**2 val_data['return_dm2'] = val_data['return_dm']**2 variance = np.var(est_data['return_dm']) print('average = ', mu) print('variance = ', variance) # Initialazing with an instance of the model object # ------------------------------------------------- hidden_dim = 12 model = rnn(1, hidden_dim, 1, variance=variance, model_type='Jordan') print('w dimensions =', model.w_dim) #state, est_data['sigma2'] = model.foward_prop(est_data['return']) # Minimize loss function # ---------------------- # Initializing with random weights uniformly on the interval -1/sqrt(k) to -1/sqrt(k) # where k = previous layers dimensions. W_H = np.random.uniform(-np.sqrt(1. / 3), np.sqrt(1. / 3), (hidden_dim, 3)) W_O = np.random.uniform(-np.sqrt(1. / hidden_dim), np.sqrt(1. / hidden_dim), (1, hidden_dim + 1)) w0 = np.hstack((W_H.flatten(), W_O.flatten())) print('Initial weights =', w0) #est_dct = json.load(open('Jordan_est_Mon_02-11-2015_06.57.json')) #w0 = np.array(est_dct['w_opt'])
import pandas as pd import numpy as np from RNNnumpy import RNNnumpy as rnn import json from pprint import pprint import scipy.stats as sps import matplotlib.pyplot as plt from pprint import pprint #est_dct = json.load(open('Jordan_est_Mon_02-11-2015_06.57.json')) # Bedste bud kl 8:00 est_dct = json.load(open('Jordan_est_Tue_03-11-2015_18.35.json')) #pprint(est_dct) model = rnn(1,est_dct['D-M-K'][1],1, #w = est_dct['w_opt'], w = est_dct['w_cross_val'], #w = est_dct['w_all'][40], variance=est_dct['variance'], mu = est_dct['mean'], model_type=est_dct['model_type']) print(model) print(est_dct['variance']) est_dct_garch = json.load(open('GARCH_est_Sun_01-11-2015_21.06.57.json')) pprint(est_dct_garch) # Plot News impact curve # ---------------------- col_lst = plt.cm.Set1(np.linspace(0, 1, 9)) y = np.arange(-4,4,0.001) sigma2 = np.array([model.forward_prop([yi]) for yi in y]) # GARCH par
import numpy as np from RNNnumpy import RNNnumpy as rnn import json from pprint import pprint import scipy.stats as sps import matplotlib.pyplot as plt from pprint import pprint #est_dct = json.load(open('Jordan_est_Mon_02-11-2015_06.57.json')) # Bedste bud kl 8:00 est_dct = json.load(open('Jordan_est_Tue_03-11-2015_18.35.json')) #pprint(est_dct) model = rnn( 1, est_dct['D-M-K'][1], 1, #w = est_dct['w_opt'], w=est_dct['w_cross_val'], #w = est_dct['w_all'][40], variance=est_dct['variance'], mu=est_dct['mean'], model_type=est_dct['model_type']) print(model) print(est_dct['variance']) est_dct_garch = json.load(open('GARCH_est_Sun_01-11-2015_21.06.57.json')) pprint(est_dct_garch) # Plot News impact curve # ---------------------- col_lst = plt.cm.Set1(np.linspace(0, 1, 9)) y = np.arange(-4, 4, 0.001) sigma2 = np.array([model.forward_prop([yi]) for yi in y])
val_data['return'] = df.loc[val_index,'return'] mu = np.mean(est_data['return']) est_data['return_dm'] = est_data['return'] - mu val_data['return_dm'] = val_data['return'] - mu est_data['return_dm2'] = est_data['return_dm']**2 val_data['return_dm2'] = val_data['return_dm']**2 variance = np.var(est_data['return_dm']) print('average = ', mu) print('variance = ', variance) # Initialazing with an instance of the model object # ------------------------------------------------- hidden_dim = 12 model = rnn(1,hidden_dim,1,variance=variance,model_type='Jordan') print('w dimensions =',model.w_dim) #state, est_data['sigma2'] = model.foward_prop(est_data['return']) # Minimize loss function # ---------------------- # Initializing with random weights uniformly on the interval -1/sqrt(k) to -1/sqrt(k) # where k = previous layers dimensions. W_H = np.random.uniform(-np.sqrt(1./3),np.sqrt(1./3),(hidden_dim,3)) W_O = np.random.uniform(-np.sqrt(1./hidden_dim),np.sqrt(1./hidden_dim),(1,hidden_dim+1)) w0 = np.hstack((W_H.flatten(),W_O.flatten())) print('Initial weights =',w0) #est_dct = json.load(open('Jordan_est_Mon_02-11-2015_06.57.json')) #w0 = np.array(est_dct['w_opt'])