How to shape input for Ragged tensor with LSTM

Question

I'm getting the following error when I try to pass multiple ragged tensors to my model:

ValueError: Layer "sequential_3" expects 1 input(s), but it received 2 input tensors

I suspect it has something to do with the "shape" argument in the Input layer of the model.

Yes, I've reviewed the ragged tensor documentation.
Yes, I've spent many hours scouring the Google-net.
Yes, I've browsed stackoverflow.com related articles.
Yes, I'm most likely noobing it up right now and there is a simple solution.

Below is a reproducible example (tensorflow 2.7; keras 2.7; python 3.7)

dependencies

import tensorflow as tf
import pandas as pd
import keras
import keras.layers

Dummy Dataset

d0 = pd.DataFrame(data={
    "id":[
        1,
        2, 2,
        3, 3, 3,
        4, 4, 4, 4,
        5, 5, 5, 5, 5,
        6, 6, 6, 6, 6, 6
    ],
    "date":[
        pd.to_datetime("2008-03-31"),
        pd.to_datetime("2008-03-31"), pd.to_datetime("2008-06-30"),
        pd.to_datetime("2008-03-31"), pd.to_datetime("2008-06-30"), pd.to_datetime("2008-09-30"),
        pd.to_datetime("2008-03-31"), pd.to_datetime("2008-06-30"), pd.to_datetime("2008-09-30"), pd.to_datetime("2008-12-31"),
        pd.to_datetime("2008-03-31"), pd.to_datetime("2008-06-30"), pd.to_datetime("2008-09-30"), pd.to_datetime("2008-12-31"), pd.to_datetime("2009-03-31"),
        pd.to_datetime("2008-03-31"), pd.to_datetime("2008-06-30"), pd.to_datetime("2008-09-30"), pd.to_datetime("2008-12-31"), pd.to_datetime("2009-03-31"), pd.to_datetime("2009-06-30")
    ],
    "date2":[
        1,
        1, 2,
        1, 2, 3,
        1, 2, 3, 4,
        1, 2, 3, 4, 5,
        1, 2, 3, 4, 5, 6
    ],
    "input":[
        10,
        11, 12,
        13, 14, 15,
        16, 17, 18, 19,
        20, 21, 22, 23, 24,
        25, 26, 27, 28, 29, 30
    ],
    "target":[
        60,
        61, 62,
        63, 64, 65,
        66, 67, 68, 69,
        70, 71, 72, 73, 74, 
        75, 76, 77, 78, 79, 80
    ],
})

inputs

inputs = []
inputs.append(tf.ragged.constant(d0.groupby("id")["input"].apply(list)))
inputs.append(tf.ragged.constant(d0.groupby("id")["date2"].apply(list)))

target

target = tf.ragged.constant(d0.groupby("id")["target"].apply(list))

model

mod1 = keras.Sequential([
    keras.layers.Input(shape=(None, 4), dtype=tf.float32, batch_size=1, ragged=True),
    keras.layers.LSTM(units=32, dtype=tf.float32, return_sequences=True, use_bias=False),
    keras.layers.Dense(units=32),
    keras.layers.Dense(units=1)
])

compile

mod1.compile(
    optimizer="adam",
    loss="mse",
    metrics=["accuracy"]
)

fit

history = mod1.fit(
    x=inputs,
    y=target,
    epochs=1
)

The error occurs during the fit step.

How to shape input for Ragged tensor with LSTM

0 Answers