Predicción del valor de una acción a tres días. Apple (LSTM)#

Importar las librería requeridas#

import tensorflow as tf

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
#%matplotlib inline
from sklearn.preprocessing import MinMaxScaler

#setting figure size
from matplotlib.pylab import rcParams
rcParams['figure.figsize'] = 20,10

#importing required libraries
from sklearn.preprocessing import MinMaxScaler

# importa objetos de keras
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Dense, Dropout, LSTM
print("Versión de Tensorflow: ", tf.__version__)

# optimizador
from tensorflow.keras.optimizers import Adam

Versión de Tensorflow:  2.9.1

Lectura de los datos#

Estos datos corresponden a la empresa Apple. Son 3019 datos que corresponden a observaciones del precio de la acción, el número de transacciones de la acción (compra-venta). Los datos son diarios (dias hábiles o comerciales). Están entre el 3 de enero de 2006 hasta el 1 de enero de 2018.

la columna Date es la fecha, Open es el valor de acción a la apertura del mercado, High el valor más alto alcanzado en el día, Low el valor más bajo del día, Close el valor al cierre, Volume es el volúmenes de acciones transadas en el día y Name es el código de identificación de la empresa, Apple en este caso.

Los datos puede ser bajados directamente de Kaggle

#reading from a local file

df = pd.read_csv('https://raw.githubusercontent.com/AprendizajeProfundo/Libro-Fundamentos/main/Redes_Recurrentes/Datos/AAPL_2006-01-01_to_2018-01-01.csv')

# looking at the first five rows of the data
print('\n Shape of the data:')
print(df.shape)
df.head()

 Shape of the data:
(3019, 7)

	Date	Open	High	Low	Close	Volume	Name
0	2006-01-03	10.34	10.68	10.32	10.68	201853036	AAPL
1	2006-01-04	10.73	10.85	10.64	10.71	155225609	AAPL
2	2006-01-05	10.69	10.70	10.54	10.63	112396081	AAPL
3	2006-01-06	10.75	10.96	10.65	10.90	176139334	AAPL
4	2006-01-09	10.96	11.03	10.82	10.86	168861224	AAPL

Vamos a cambiar el índice de los datos. Tomaremos la fecha como indice: df.index. Los datos se reordenan para invertir la tabla, debido a que los datos contienen las observaciones más recientes en la parte superior de la tabla.

Extrae datos para la serie que se desea predecir-close#

#creating dataframe with date and the target variable

df['Date'] = pd.to_datetime(df.Date,format='%Y-%m-%d')
df.index = df['Date']
# df = df.sort_index(ascending=True, axis=0)
data = pd.DataFrame(df[['Date', 'Close']])
#
#setting index
data.index = data.Date
data.drop('Date', axis=1, inplace=True)
data.head()

	Close
Date
2006-01-03	10.68
2006-01-04	10.71
2006-01-05	10.63
2006-01-06	10.90
2006-01-09	10.86

Visualización de la serie precio al cierre#

# plot
len_data = len(data)
len_train = int(len_data*0.8) # 80%  = 3019
len_test = len_data- len_train # 20% = 2415
print (len_data, '=', len_train, '+',len_test)

3019 = 2415 + 604

plt.figure(figsize=(16,8))
plt.plot(data[:len_train], label='Conjunto de entrenamiento (Training set): {} puntos (80%)'.format(len_train))
plt.plot(data['Close'][len_train:], label='Conjunto de validación (Validation set): {} puntos (20%)'.format(len_test)) #248 data
plt.title("Apple: Historia del precio la acción al cierre (Close)", size = 20)
plt.legend()
plt.show()

../../_images/rnr_accion_Apple_Prediccion_tres_dias-dropout_12_0.png

Preparación de los datos para el entrenamiento de la red LSTM#

Para evitar problemas con las tendencias y para mejorar la estimación (entrenamiento) los datos se van a transformar a la escala \([0,1]\). Para las predicciones se utiliza la transformación inversa.

Primero extrae los valores y se crea el objeto MinMaxScaler#

#creating train and test sets
dataset = data.values


# create the scaler object and scale the data
scaler = MinMaxScaler(feature_range=(0, 1))
#scaled_data = np.array(scaler.fit_transform(dataset))
dataset = np.squeeze(np.array(scaler.fit_transform(dataset)),axis=1)
# dataset = pd.DataFrame(scaled_data,index=data.index, columns=['serie'])

dataset.shape

(3019,)

Crea datos de entrenamiento#

La red LSTM tendrá como entrada «time_step» datos consecutivos, y como salida 5 datos (la predicción a partir de esos «time_step» datos se hace para los siguentes 5 días). Se conformará de esta forma el set de entrenamiento

Número de datos consecutivos para entrenamiento: time_step = 60.
Días a predecir: days = 1

Función para crear los datos de entrenamiento#

def univariate_data(dataset, start_index, end_index, history_size, target_size):
    ''' dataset: conjunto de datos
        start_index: índice inicial de donde empezar a tomar los datos
        end_index: índice final para tomar los datos. None para tomarlos todos
        history_size: tamaño de la venytana para crear las secuencias
        target_size: dentro de cuántas observaciones futuras desea pronosticar
    '''  
    data = []
    labels = []

    start_index = start_index + history_size
    if end_index is None:
        end_index = len(dataset) - target_size

    for i in range(start_index, end_index):
        indices = range(i-history_size, i)
        # Reshape data from (history_size,) to (history_size, 1)
        data.append(np.reshape(dataset[indices], (history_size, 1)))
        labels.append(dataset[i+target_size])

    return np.array(data), np.array(labels)

Se coloca una semilla para gerantizar reproductibidad

tf.random.set_seed(100)

#
# hiperparámetros para crear las secuencias
past_history = 60 # tamaño secuencias de entrada
future_target = 3 # días adelante
TRAIN_SPLIT = int(len_data*0.8) #2415: nuḿer0 de datos entreno

# secuencias de entrenamiento
X_train, y_train = univariate_data(dataset, 0, TRAIN_SPLIT,
                                           past_history,
                                           future_target)
#
#secuencias de validación
# No se usará ningún dato que el modelo haya visto
X_test, y_test = univariate_data(dataset, TRAIN_SPLIT, None,
                                       past_history,
                                       future_target)

print(TRAIN_SPLIT)
print(X_train.shape)
print(y_train.shape)
print(X_test.shape)
print(y_test.shape)

2415
(2355, 60, 1)
(2355,)
(541, 60, 1)
(541,)

Crea el modelo LSTM#

Omitimos esta sección. Usaremso el modelo entrenado a un día

# shapes
input_shape = (X_train.shape[1], X_train.shape[2])
units = 64

# layers
inputs = Input(input_shape)
#x = Dropout(0.0, name= 'Dropout_01')(inputs)
#x = LSTM(units=units, name='LSTM_layer')(x)
x = LSTM(units=units, return_sequences=True,name='LSTM_layer')(inputs)
x = Dropout(0.4, name= 'Dropout_02')(x)
x = LSTM(units=units//2, name='LSTM_layer_2')(x)
x = Dropout(0.4, name= 'Dropout_03')(x)
outputs = Dense(1)(x)

# model
model_01 = Model(inputs=inputs, outputs=outputs, name='series_LSTM_model')
model_01.summary()

Model: "series_LSTM_model"

_________________________________________________________________

 Layer (type)                Output Shape              Param #

=================================================================

 input_1 (InputLayer)        [(None, 60, 1)]           0

 LSTM_layer (LSTM)           (None, 60, 64)            16896

 Dropout_02 (Dropout)        (None, 60, 64)            0

 LSTM_layer_2 (LSTM)         (None, 32)                12416

 Dropout_03 (Dropout)        (None, 32)                0

 dense (Dense)               (None, 1)                 33

=================================================================

Total params: 29,345

Trainable params: 29,345

Non-trainable params: 0

_________________________________________________________________

Compila#

Se usará el optimizador Adam y la función de pérdida MSE

model_01.compile(loss='mean_squared_error',
  optimizer=Adam(0.001))

Entrena el modelo#

#history = model_01.fit(X_train,y_train,epochs=20,batch_size=32)
tf.random.set_seed(100)
history = model_01.fit(
    X_train, y_train,
    epochs=15,
    batch_size=32,
    validation_split=0.1,
    verbose=1,
    shuffle=False
)

Epoch 1/15

 1/67 [..............................] - ETA: 7:44 - loss: 2.4157e-04


 2/67 [..............................] - ETA: 5s - loss: 1.4562e-04


 3/67 [>.............................] - ETA: 4s - loss: 1.2524e-04


 4/67 [>.............................] - ETA: 4s - loss: 1.0101e-04


 5/67 [=>............................] - ETA: 4s - loss: 9.9967e-05


 6/67 [=>............................] - ETA: 4s - loss: 1.1648e-04


 7/67 [==>...........................] - ETA: 3s - loss: 1.1462e-04


 8/67 [==>...........................] - ETA: 3s - loss: 1.0868e-04


 9/67 [===>..........................] - ETA: 3s - loss: 1.1215e-04


10/67 [===>..........................] - ETA: 3s - loss: 1.2017e-04


11/67 [===>..........................] - ETA: 3s - loss: 1.3608e-04


12/67 [====>.........................] - ETA: 3s - loss: 1.5585e-04


13/67 [====>.........................] - ETA: 3s - loss: 2.0753e-04


14/67 [=====>........................] - ETA: 3s - loss: 2.1871e-04


15/67 [=====>........................] - ETA: 3s - loss: 3.1331e-04


16/67 [======>.......................] - ETA: 3s - loss: 3.2149e-04


18/67 [=======>......................] - ETA: 3s - loss: 3.8504e-04


19/67 [=======>......................] - ETA: 3s - loss: 3.9329e-04


20/67 [=======>......................] - ETA: 3s - loss: 4.2395e-04


21/67 [========>.....................] - ETA: 2s - loss: 4.5621e-04


22/67 [========>.....................] - ETA: 2s - loss: 4.9952e-04


23/67 [=========>....................] - ETA: 2s - loss: 5.0137e-04


24/67 [=========>....................] - ETA: 2s - loss: 4.8637e-04


25/67 [==========>...................] - ETA: 2s - loss: 4.8661e-04


26/67 [==========>...................] - ETA: 2s - loss: 5.1376e-04


27/67 [===========>..................] - ETA: 2s - loss: 6.0306e-04


28/67 [===========>..................] - ETA: 2s - loss: 7.3945e-04


29/67 [===========>..................] - ETA: 2s - loss: 8.4674e-04


30/67 [============>.................] - ETA: 2s - loss: 9.2930e-04


31/67 [============>.................] - ETA: 2s - loss: 9.6262e-04


32/67 [=============>................] - ETA: 2s - loss: 0.0010


33/67 [=============>................] - ETA: 2s - loss: 0.0011


34/67 [==============>...............] - ETA: 2s - loss: 0.0011


35/67 [==============>...............] - ETA: 2s - loss: 0.0011


36/67 [===============>..............] - ETA: 1s - loss: 0.0011


37/67 [===============>..............] - ETA: 1s - loss: 0.0012


38/67 [================>.............] - ETA: 1s - loss: 0.0012


39/67 [================>.............] - ETA: 1s - loss: 0.0013


40/67 [================>.............] - ETA: 1s - loss: 0.0013


41/67 [=================>............] - ETA: 1s - loss: 0.0013


42/67 [=================>............] - ETA: 1s - loss: 0.0014


43/67 [==================>...........] - ETA: 1s - loss: 0.0015


44/67 [==================>...........] - ETA: 1s - loss: 0.0015


45/67 [===================>..........] - ETA: 1s - loss: 0.0016


46/67 [===================>..........] - ETA: 1s - loss: 0.0016


47/67 [====================>.........] - ETA: 1s - loss: 0.0017


48/67 [====================>.........] - ETA: 1s - loss: 0.0018


49/67 [====================>.........] - ETA: 1s - loss: 0.0019


50/67 [=====================>........] - ETA: 1s - loss: 0.0020


51/67 [=====================>........] - ETA: 1s - loss: 0.0022


52/67 [======================>.......] - ETA: 0s - loss: 0.0023


53/67 [======================>.......] - ETA: 0s - loss: 0.0026


54/67 [=======================>......] - ETA: 0s - loss: 0.0026


55/67 [=======================>......] - ETA: 0s - loss: 0.0027


56/67 [========================>.....] - ETA: 0s - loss: 0.0027


57/67 [========================>.....] - ETA: 0s - loss: 0.0027


58/67 [========================>.....] - ETA: 0s - loss: 0.0027


60/67 [=========================>....] - ETA: 0s - loss: 0.0030


61/67 [==========================>...] - ETA: 0s - loss: 0.0032


62/67 [==========================>...] - ETA: 0s - loss: 0.0033


63/67 [===========================>..] - ETA: 0s - loss: 0.0034


64/67 [===========================>..] - ETA: 0s - loss: 0.0034


65/67 [============================>.] - ETA: 0s - loss: 0.0035


66/67 [============================>.] - ETA: 0s - loss: 0.0036


67/67 [==============================] - ETA: 0s - loss: 0.0036


67/67 [==============================] - 13s 98ms/step - loss: 0.0036 - val_loss: 0.0017

Epoch 2/15

 1/67 [..............................] - ETA: 5s - loss: 0.0616


 2/67 [..............................] - ETA: 4s - loss: 0.0603


 3/67 [>.............................] - ETA: 4s - loss: 0.0552


 4/67 [>.............................] - ETA: 3s - loss: 0.0490


 5/67 [=>............................] - ETA: 3s - loss: 0.0435


 6/67 [=>............................] - ETA: 3s - loss: 0.0383


 7/67 [==>...........................] - ETA: 3s - loss: 0.0339


 8/67 [==>...........................] - ETA: 3s - loss: 0.0300


 9/67 [===>..........................] - ETA: 3s - loss: 0.0268


10/67 [===>..........................] - ETA: 3s - loss: 0.0242


11/67 [===>..........................] - ETA: 3s - loss: 0.0220


12/67 [====>.........................] - ETA: 3s - loss: 0.0204


13/67 [====>.........................] - ETA: 3s - loss: 0.0193


14/67 [=====>........................] - ETA: 3s - loss: 0.0184


15/67 [=====>........................] - ETA: 3s - loss: 0.0173


16/67 [======>.......................] - ETA: 3s - loss: 0.0165


17/67 [======>.......................] - ETA: 3s - loss: 0.0159


18/67 [=======>......................] - ETA: 3s - loss: 0.0153


19/67 [=======>......................] - ETA: 3s - loss: 0.0148


20/67 [=======>......................] - ETA: 2s - loss: 0.0141


21/67 [========>.....................] - ETA: 2s - loss: 0.0135


22/67 [========>.....................] - ETA: 2s - loss: 0.0129


23/67 [=========>....................] - ETA: 2s - loss: 0.0123


24/67 [=========>....................] - ETA: 2s - loss: 0.0118


25/67 [==========>...................] - ETA: 2s - loss: 0.0114


26/67 [==========>...................] - ETA: 2s - loss: 0.0110


27/67 [===========>..................] - ETA: 2s - loss: 0.0106


28/67 [===========>..................] - ETA: 2s - loss: 0.0103


29/67 [===========>..................] - ETA: 2s - loss: 0.0100


30/67 [============>.................] - ETA: 2s - loss: 0.0098


31/67 [============>.................] - ETA: 2s - loss: 0.0095


32/67 [=============>................] - ETA: 2s - loss: 0.0094


33/67 [=============>................] - ETA: 2s - loss: 0.0092


34/67 [==============>...............] - ETA: 2s - loss: 0.0090


35/67 [==============>...............] - ETA: 2s - loss: 0.0089


36/67 [===============>..............] - ETA: 2s - loss: 0.0087


37/67 [===============>..............] - ETA: 2s - loss: 0.0086


38/67 [================>.............] - ETA: 2s - loss: 0.0085


39/67 [================>.............] - ETA: 2s - loss: 0.0084


40/67 [================>.............] - ETA: 2s - loss: 0.0082


41/67 [=================>............] - ETA: 2s - loss: 0.0081


42/67 [=================>............] - ETA: 2s - loss: 0.0080


43/67 [==================>...........] - ETA: 2s - loss: 0.0078


44/67 [==================>...........] - ETA: 1s - loss: 0.0078


45/67 [===================>..........] - ETA: 1s - loss: 0.0077


46/67 [===================>..........] - ETA: 1s - loss: 0.0076


47/67 [====================>.........] - ETA: 1s - loss: 0.0076


48/67 [====================>.........] - ETA: 1s - loss: 0.0077


49/67 [====================>.........] - ETA: 1s - loss: 0.0077


50/67 [=====================>........] - ETA: 1s - loss: 0.0077


51/67 [=====================>........] - ETA: 1s - loss: 0.0078


52/67 [======================>.......] - ETA: 1s - loss: 0.0078


53/67 [======================>.......] - ETA: 1s - loss: 0.0079


54/67 [=======================>......] - ETA: 1s - loss: 0.0080


55/67 [=======================>......] - ETA: 1s - loss: 0.0079


56/67 [========================>.....] - ETA: 0s - loss: 0.0079


57/67 [========================>.....] - ETA: 0s - loss: 0.0078


58/67 [========================>.....] - ETA: 0s - loss: 0.0078


59/67 [=========================>....] - ETA: 0s - loss: 0.0077


60/67 [=========================>....] - ETA: 0s - loss: 0.0078


61/67 [==========================>...] - ETA: 0s - loss: 0.0079


62/67 [==========================>...] - ETA: 0s - loss: 0.0081


63/67 [===========================>..] - ETA: 0s - loss: 0.0081


64/67 [===========================>..] - ETA: 0s - loss: 0.0082


65/67 [============================>.] - ETA: 0s - loss: 0.0083


66/67 [============================>.] - ETA: 0s - loss: 0.0083


67/67 [==============================] - ETA: 0s - loss: 0.0083


67/67 [==============================] - 6s 91ms/step - loss: 0.0083 - val_loss: 0.0022

Epoch 3/15

 1/67 [..............................] - ETA: 7s - loss: 0.0553


 2/67 [..............................] - ETA: 6s - loss: 0.0567


 3/67 [>.............................] - ETA: 6s - loss: 0.0547


 4/67 [>.............................] - ETA: 5s - loss: 0.0518


 5/67 [=>............................] - ETA: 5s - loss: 0.0475


 6/67 [=>............................] - ETA: 5s - loss: 0.0439


 7/67 [==>...........................] - ETA: 5s - loss: 0.0396


 8/67 [==>...........................] - ETA: 5s - loss: 0.0357


 9/67 [===>..........................] - ETA: 5s - loss: 0.0322


10/67 [===>..........................] - ETA: 5s - loss: 0.0292


11/67 [===>..........................] - ETA: 5s - loss: 0.0266


12/67 [====>.........................] - ETA: 5s - loss: 0.0244


13/67 [====>.........................] - ETA: 5s - loss: 0.0226


14/67 [=====>........................] - ETA: 5s - loss: 0.0211


15/67 [=====>........................] - ETA: 5s - loss: 0.0197


16/67 [======>.......................] - ETA: 5s - loss: 0.0186


17/67 [======>.......................] - ETA: 5s - loss: 0.0177


18/67 [=======>......................] - ETA: 5s - loss: 0.0169


19/67 [=======>......................] - ETA: 5s - loss: 0.0162


20/67 [=======>......................] - ETA: 4s - loss: 0.0154


21/67 [========>.....................] - ETA: 4s - loss: 0.0147


22/67 [========>.....................] - ETA: 4s - loss: 0.0141


23/67 [=========>....................] - ETA: 4s - loss: 0.0135


24/67 [=========>....................] - ETA: 4s - loss: 0.0130


25/67 [==========>...................] - ETA: 4s - loss: 0.0126


26/67 [==========>...................] - ETA: 4s - loss: 0.0122


27/67 [===========>..................] - ETA: 4s - loss: 0.0118


28/67 [===========>..................] - ETA: 4s - loss: 0.0115


29/67 [===========>..................] - ETA: 4s - loss: 0.0112


30/67 [============>.................] - ETA: 4s - loss: 0.0109


31/67 [============>.................] - ETA: 3s - loss: 0.0106


32/67 [=============>................] - ETA: 3s - loss: 0.0104


33/67 [=============>................] - ETA: 3s - loss: 0.0102


34/67 [==============>...............] - ETA: 3s - loss: 0.0100


35/67 [==============>...............] - ETA: 3s - loss: 0.0097


36/67 [===============>..............] - ETA: 3s - loss: 0.0096


37/67 [===============>..............] - ETA: 3s - loss: 0.0095


38/67 [================>.............] - ETA: 3s - loss: 0.0093


39/67 [================>.............] - ETA: 3s - loss: 0.0091


40/67 [================>.............] - ETA: 3s - loss: 0.0090


41/67 [=================>............] - ETA: 2s - loss: 0.0088


42/67 [=================>............] - ETA: 2s - loss: 0.0086


43/67 [==================>...........] - ETA: 2s - loss: 0.0085


44/67 [==================>...........] - ETA: 2s - loss: 0.0083


45/67 [===================>..........] - ETA: 2s - loss: 0.0082


46/67 [===================>..........] - ETA: 2s - loss: 0.0081


47/67 [====================>.........] - ETA: 2s - loss: 0.0081


48/67 [====================>.........] - ETA: 2s - loss: 0.0081


49/67 [====================>.........] - ETA: 1s - loss: 0.0080


50/67 [=====================>........] - ETA: 1s - loss: 0.0080


51/67 [=====================>........] - ETA: 1s - loss: 0.0080


52/67 [======================>.......] - ETA: 1s - loss: 0.0081


53/67 [======================>.......] - ETA: 1s - loss: 0.0081


54/67 [=======================>......] - ETA: 1s - loss: 0.0081


55/67 [=======================>......] - ETA: 1s - loss: 0.0081


56/67 [========================>.....] - ETA: 1s - loss: 0.0080


57/67 [========================>.....] - ETA: 1s - loss: 0.0079


58/67 [========================>.....] - ETA: 0s - loss: 0.0078


59/67 [=========================>....] - ETA: 0s - loss: 0.0078


60/67 [=========================>....] - ETA: 0s - loss: 0.0077


61/67 [==========================>...] - ETA: 0s - loss: 0.0078


62/67 [==========================>...] - ETA: 0s - loss: 0.0078


63/67 [===========================>..] - ETA: 0s - loss: 0.0078


64/67 [===========================>..] - ETA: 0s - loss: 0.0079


65/67 [============================>.] - ETA: 0s - loss: 0.0079


66/67 [============================>.] - ETA: 0s - loss: 0.0079


67/67 [==============================] - ETA: 0s - loss: 0.0080


67/67 [==============================] - 7s 106ms/step - loss: 0.0080 - val_loss: 0.0045

Epoch 4/15

 1/67 [..............................] - ETA: 6s - loss: 0.0180


 2/67 [..............................] - ETA: 5s - loss: 0.0195


 3/67 [>.............................] - ETA: 5s - loss: 0.0199


 4/67 [>.............................] - ETA: 5s - loss: 0.0194


 5/67 [=>............................] - ETA: 5s - loss: 0.0189


 6/67 [=>............................] - ETA: 4s - loss: 0.0180


 7/67 [==>...........................] - ETA: 4s - loss: 0.0170


 8/67 [==>...........................] - ETA: 4s - loss: 0.0158


 9/67 [===>..........................] - ETA: 4s - loss: 0.0145


10/67 [===>..........................] - ETA: 4s - loss: 0.0134


11/67 [===>..........................] - ETA: 4s - loss: 0.0123


12/67 [====>.........................] - ETA: 4s - loss: 0.0113


13/67 [====>.........................] - ETA: 4s - loss: 0.0105


14/67 [=====>........................] - ETA: 4s - loss: 0.0098


15/67 [=====>........................] - ETA: 3s - loss: 0.0092


16/67 [======>.......................] - ETA: 3s - loss: 0.0087


17/67 [======>.......................] - ETA: 3s - loss: 0.0083


18/67 [=======>......................] - ETA: 3s - loss: 0.0079


19/67 [=======>......................] - ETA: 3s - loss: 0.0076


20/67 [=======>......................] - ETA: 3s - loss: 0.0072


21/67 [========>.....................] - ETA: 3s - loss: 0.0071


22/67 [========>.....................] - ETA: 3s - loss: 0.0069


23/67 [=========>....................] - ETA: 3s - loss: 0.0067


24/67 [=========>....................] - ETA: 3s - loss: 0.0066


25/67 [==========>...................] - ETA: 3s - loss: 0.0064


26/67 [==========>...................] - ETA: 2s - loss: 0.0062


27/67 [===========>..................] - ETA: 2s - loss: 0.0060


28/67 [===========>..................] - ETA: 2s - loss: 0.0058


29/67 [===========>..................] - ETA: 2s - loss: 0.0056


30/67 [============>.................] - ETA: 2s - loss: 0.0054


31/67 [============>.................] - ETA: 2s - loss: 0.0053


32/67 [=============>................] - ETA: 2s - loss: 0.0051


33/67 [=============>................] - ETA: 2s - loss: 0.0050


34/67 [==============>...............] - ETA: 2s - loss: 0.0049


35/67 [==============>...............] - ETA: 2s - loss: 0.0048


36/67 [===============>..............] - ETA: 2s - loss: 0.0047


37/67 [===============>..............] - ETA: 2s - loss: 0.0046


39/67 [================>.............] - ETA: 1s - loss: 0.0044


40/67 [================>.............] - ETA: 1s - loss: 0.0043


41/67 [=================>............] - ETA: 1s - loss: 0.0043


42/67 [=================>............] - ETA: 1s - loss: 0.0042


43/67 [==================>...........] - ETA: 1s - loss: 0.0042


44/67 [==================>...........] - ETA: 1s - loss: 0.0041


45/67 [===================>..........] - ETA: 1s - loss: 0.0041


46/67 [===================>..........] - ETA: 1s - loss: 0.0040

plt.plot(history.history['loss'], label='train')
plt.plot(history.history['val_loss'], label='test')
plt.legend();

../../_images/rnr_accion_Apple_Prediccion_tres_dias-dropout_33_0.png

Predicciones#

Prepara los datos de validación#

X_test.shape

(541, 60, 1)

Calcula predicciones#

# predictions
prediction = model_01.predict(X_test)
#prediction = scaler.inverse_transform(prediction)

17/17 [==============================] - 2s 18ms/step

print(prediction.shape)
print(y_test.shape)

(541, 1)
(541,)

Elimina dimensiones sobrante para los gráficos#

y_train_p = y_train# np.squeeze(y_train, axis=-1)
y_test_p = y_test #np.squeeze(y_test, axis=-1)
y_pred_p = np.squeeze(prediction, axis=-1)

print(y_train_p.shape)
print(y_test_p.shape)
print(y_pred_p.shape)

k=0
for i,j in zip(y_test_p, y_pred_p):
    print (i,j, i-j)
    k+=1
    if k==10:
        break

(2355,)
(541,)
(541,)
6727745596406196 0.64994586 0.022828704499933572
6698782361981322 0.6543955 0.015482715511028977
6474169523584349 0.6584121 -0.010995146526147637
6435157820073296 0.6614015 -0.017885728231317843
6412105449816764 0.66360754 -0.022396992764753043
6212318240926825 0.664968 -0.04373618967074522
6321078141624307 0.664171 -0.03206316629106054
6292114907199433 0.66194737 -0.03273587885555718
6504906017259725 0.65891695 -0.008426348499857572
6592977893368012 0.6542779 0.00501986861383974

Gráfica de las predicciones#

plt.plot(np.arange(0, len(y_train_p)), y_train_p, 'g', label="historia")
plt.plot(np.arange(len(y_train_p), len(y_train_p) + len(y_test_p)), y_test_p, marker='.', label="verdadero")
plt.plot(np.arange(len(y_train_p), len(y_train_p) + len(y_test_p)), y_pred_p, 'r', label="predicción")
#plt.ylabel('Valor')
plt.xlabel('Time Step')
plt.title("Apple: Historia del precio la acción al cierre. Escala (0,1)", size = 20)
plt.legend()
plt.show();

../../_images/rnr_accion_Apple_Prediccion_tres_dias-dropout_44_0.png

Regreso a la escala original#

y_pred_or = scaler.inverse_transform(y_pred_p.reshape(-1,1))
y_test_or = scaler.inverse_transform(y_test_p.reshape(-1,1))

k=0
for i,j in zip(y_test_or, y_pred_or):
    print (i,j, i-j)
    k+=1
    if k==10:
        break

[121.06] [117.19784] [3.86216217]
[120.57] [117.95063] [2.61936981]
[116.77] [118.63016] [-1.86015747]
[116.11] [119.1359] [-3.0259024]
[115.72] [119.509125] [-3.78912476]
[112.34] [119.73929] [-7.39928833]
[114.18] [119.60445] [-5.42444641]
[113.69] [119.22826] [-5.53825623]
[117.29] [118.71557] [-1.42556854]
[118.78] [117.93074] [0.84925964]

rmsLSTM = np.sqrt(np.mean(np.power(y_pred_or-y_test_or,2)))
print(rmsLSTM )

8.170484800047866

plt.plot(np.arange(0, len(y_test_or)), y_test_or, marker='.', label="verdadero")
plt.plot(np.arange(0, len(y_test_or)), y_pred_or, marker='+', label="predicho")
plt.xlabel('Time Step')
plt.annotate("rms = "+str(round(rmsLSTM,2)) , xy=(100, 140), size = 15)
plt.annotate("modelo = LSTM(50), timestep=60" , xy=(100, 146), size = 15)
plt.annotate("epochs=40" , xy=(100, 143), size = 15)
plt.title("Apple: Intervalo de predicción a tres días. Escala original", size = 20)
plt.legend()
plt.show();

../../_images/rnr_accion_Apple_Prediccion_tres_dias-dropout_49_0.png

Guarda el modelo entrenado#

Python

model_01.save('../Datos/modelo_Apple_3_dia.h5')

Intervalos de confianza. TO DO#

model_01.get_config()

{'name': 'series_LSTM_model',
 'layers': [{'class_name': 'InputLayer',
   'config': {'batch_input_shape': (None, 60, 1),
    'dtype': 'float32',
    'sparse': False,
    'ragged': False,
    'name': 'input_1'},
   'name': 'input_1',
   'inbound_nodes': []},
  {'class_name': 'LSTM',
   'config': {'name': 'LSTM_layer',
    'trainable': True,
    'dtype': 'float32',
    'return_sequences': True,
    'return_state': False,
    'go_backwards': False,
    'stateful': False,
    'unroll': False,
    'time_major': False,
    'units': 64,
    'activation': 'tanh',
    'recurrent_activation': 'sigmoid',
    'use_bias': True,
    'kernel_initializer': {'class_name': 'GlorotUniform',
     'config': {'seed': None},
     'shared_object_id': 1},
    'recurrent_initializer': {'class_name': 'Orthogonal',
     'config': {'gain': 1.0, 'seed': None},
     'shared_object_id': 2},
    'bias_initializer': {'class_name': 'Zeros',
     'config': {},
     'shared_object_id': 3},
    'unit_forget_bias': True,
    'kernel_regularizer': None,
    'recurrent_regularizer': None,
    'bias_regularizer': None,
    'activity_regularizer': None,
    'kernel_constraint': None,
    'recurrent_constraint': None,
    'bias_constraint': None,
    'dropout': 0.0,
    'recurrent_dropout': 0.0,
    'implementation': 2},
   'name': 'LSTM_layer',
   'inbound_nodes': [[['input_1', 0, 0, {}]]]},
  {'class_name': 'Dropout',
   'config': {'name': 'Dropout_02',
    'trainable': True,
    'dtype': 'float32',
    'rate': 0.4,
    'noise_shape': None,
    'seed': None},
   'name': 'Dropout_02',
   'inbound_nodes': [[['LSTM_layer', 0, 0, {}]]]},
  {'class_name': 'LSTM',
   'config': {'name': 'LSTM_layer_2',
    'trainable': True,
    'dtype': 'float32',
    'return_sequences': False,
    'return_state': False,
    'go_backwards': False,
    'stateful': False,
    'unroll': False,
    'time_major': False,
    'units': 32,
    'activation': 'tanh',
    'recurrent_activation': 'sigmoid',
    'use_bias': True,
    'kernel_initializer': {'class_name': 'GlorotUniform',
     'config': {'seed': None},
     'shared_object_id': 7},
    'recurrent_initializer': {'class_name': 'Orthogonal',
     'config': {'gain': 1.0, 'seed': None},
     'shared_object_id': 8},
    'bias_initializer': {'class_name': 'Zeros',
     'config': {},
     'shared_object_id': 9},
    'unit_forget_bias': True,
    'kernel_regularizer': None,
    'recurrent_regularizer': None,
    'bias_regularizer': None,
    'activity_regularizer': None,
    'kernel_constraint': None,
    'recurrent_constraint': None,
    'bias_constraint': None,
    'dropout': 0.0,
    'recurrent_dropout': 0.0,
    'implementation': 2},
   'name': 'LSTM_layer_2',
   'inbound_nodes': [[['Dropout_02', 0, 0, {}]]]},
  {'class_name': 'Dropout',
   'config': {'name': 'Dropout_03',
    'trainable': True,
    'dtype': 'float32',
    'rate': 0.4,
    'noise_shape': None,
    'seed': None},
   'name': 'Dropout_03',
   'inbound_nodes': [[['LSTM_layer_2', 0, 0, {}]]]},
  {'class_name': 'Dense',
   'config': {'name': 'dense',
    'trainable': True,
    'dtype': 'float32',
    'units': 1,
    'activation': 'linear',
    'use_bias': True,
    'kernel_initializer': {'class_name': 'GlorotUniform',
     'config': {'seed': None}},
    'bias_initializer': {'class_name': 'Zeros', 'config': {}},
    'kernel_regularizer': None,
    'bias_regularizer': None,
    'activity_regularizer': None,
    'kernel_constraint': None,
    'bias_constraint': None},
   'name': 'dense',
   'inbound_nodes': [[['Dropout_03', 0, 0, {}]]]}],
 'input_layers': [['input_1', 0, 0]],
 'output_layers': [['dense', 0, 0]]}

Referencias#

Introducción a Redes LSTM
Time Series Forecasting with LSTMs using TensorFlow 2 and Keras in Python
Dive into Deep Learnig
Understanding LSTM Networks
Ralf C. Staudemeyer and Eric Rothstein Morris,Understanding LSTM a tutorial into Long Short-Term Memory Recurrent Neural Networks, arxiv, September 2019
Karpathy, The Unreasonable Effectiveness of Recurrent Neural Networks
Anton Lucanus, Making Automation More Efficient by Learning from Historical Trade Data, 8:43 AM, January 7, 2020
https://www.analyticsvidhya.com/blog/2018/10/predicting-stock-price-machine-learningnd-deep-learning-techniques-python/
https://www.youtube.com/watch?v=2BrpKpWwT2A&list=PLQVvvaa0QuDcOdF96TBtRtuQksErCEBYZ&index=1
https://towardsdatascience.com/using-lstms-for-stock-market-predictions-tensorflow-9e83999d4653
https://github.com/llSourcell/Reinforcement_Learning_for_Stock_Prediction/blob/master/README.md

Fundamentos de IA y AP

Predicción del valor de una acción a tres días. Apple (LSTM)

Contents