Input Shape Keras RNN

Question

I'm working with a time-series data, that has shape of 2000x1001 , where 2000 is the number of cases, 1000 rows represent the data in time-domain, displacements in X direction during 1 sec period, meaning that the timestep is 0.001. The last column represents the speed, the output value that I need to predict based on the displacements during 1 sec. How the Input Data should be shaped for RNN in Keras ? I've gone trough some tutorials, but still I'm cofused about Input Shape in RNN. Thanks in advance

#load data training data
dataset=loadtxt("Data.csv", delimiter=",")
x = dataset[:,:1000]
y = dataset[:,1000]


#Create train and test dataset with an 80:20 split
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2) 

#input scaling
scaler = StandardScaler()
x_train_s =scaler.fit_transform(x_train)
x_test_s = scaler.transform(x_test)

num_samples = x_train_s.shape[0] ## Number of samples
num_vals    = x_train_s.shape[1] # Number of elements in each sample

x_train_s = np.reshape(x_train_s, (num_samples, num_vals, 1))

#create model
model = Sequential()
model.add(LSTM(100, input_shape=(num_vals, 1)))
model.add(Dense(1, activation='relu'))
model.compile(loss='mae', optimizer='adam',metrics = ['mape'])
model.summary()

#training
history = model.fit(x_train_s, y_train,epochs=10, verbose = 1, batch_size =64)

Answer 1

look at this code: it is trying to predict next 4 values based on previous 6 values. follow the comments and see how very simple input is manipulated for using it as input in rnn/lstm

follow the comments within code

from __future__ import absolute_import, division, print_function, unicode_literals
import tensorflow as tf
from tensorflow.keras import Model
import numpy as np
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import RNN, LSTM

"""
creating a toy dataset
lets use this below ```input_sequence``` as the sequence to make data points.
as per the question, we will use 6 points to predict next 4 points
"""
input_sequence = [1,2,3,4,5,6,7,8,9,10,1,2,3,4,5,6,7,8,9,10,1,2,3,4,5,6,7,8,9,10]

X_train = []
y_train = []


**#first 6 points will be our input data points and next 4 points will be data label.
#so on we will shift by 1 and make such data points and label pairs**


for i in range(len(input_sequence)-9):
    X_train.append(input_sequence[i:i+6])
    y_train.append(input_sequence[i+6:i+10])

X_train = np.array(X_train, dtype=np.float32)
y_train = np.array(y_train, dtype=np.int32)))


**#X_test for the predictions (contains 6 points)**


X_test = np.array([[8,9,10,1,2,3]],dtype=np.float32)
print(X_train.shape)
print(y_train.shape)
print(X_test.shape)


**#we will be using basic LSTM, which accepts input in ```[num_inputs, time_steps, data_points], therefore reshaping as per that```** 
# so here:
# 1. num_inputs = how many sequence of 6 points you want to use i.e. rows (we use X_train.shape[0] )

# 2. time_steps = batches you can considered i.e. if you want to use 1 or 2 or 3 rows

# 3. data_points = number of points (for ex. in our case its 6 points we are using)

X_train = np.reshape(X_train, (X_train.shape[0], 1, X_train.shape[1]))
X_test = np.reshape(X_test, (X_test.shape[0], 1, X_test.shape[1]))
print(X_train.shape)
print(y_train.shape)
print(X_test.shape)

x_points = X_train.shape[-1]
print("one input contains {} points".format(x_points))

model = Sequential()
model.add(LSTM(4, input_shape=(1, x_points)))
model.add(Dense(4))
model.compile(loss='mean_squared_error', optimizer='adam')
model.summary()

model.fit(X_train, y_train, epochs=500, batch_size=5, verbose=2)
output = list(map(np.ceil, model.predict(X_test)))
print(output)

hope it helps. ask for any doubt pls.

Answer 2

Like explained in the doc , Keras expects the following shape for a RNN:

(batch_size, timesteps, input_dim)

• batch_size is the umber of samples you feed before a backprop
• timesteps is the number of timesteps for each sample
• input_dim is the number of features for each timestep

EDIT more details:

In your case you should go for

batch_input_shape = (batch_size, timesteps, 1)

With batch_size and timesteps selected as you wish.

What about the timesteps?

Let's say you take one of your 2000 samples, and let's say that your sample has 10 elements instead of 1000, for example:

[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]

Then, if we chose timesteps=3 , then you get a batch of length 8:

[[[0], [1], [2]],
 [[1], [2], [3]],
 [[2], [3], [4]],
 [[3], [4], [5]],
 [[4], [5], [6]],
 [[5], [6], [7]],
 [[6], [7], [8]],
 [[7], [8], [9]]]