Keras 仅在使用验证生成器时卡在第一个时期

Question

I am reading a book on deep learning named Deep learning with python . The book is old in terms of code but I read the official documentation to get through it. Anyways

This is a program that is supposed to train a simple model for time series prediction of temperature using dataset available here .

The program goes like this

import numpy as np
from keras.models import Sequential
from keras import layers
from keras.optimizers import RMSprop

#Loading file 
f = open(fname) # fname is the filepath for the csv file
data = f.read()
f.close()
lines = data.split('\n')
header = lines[0].split(',')
lines = lines[1:]

# Converting into numpy array
float_data = np.zeros((len(lines), len(header) - 1))
for i, line in enumerate(lines):
values = [float(x) for x in line.split(',')[1:]]
float_data[i, :] = values

# Normalizing the data 
mean = float_data[:200000].mean(axis=0)
float_data -= mean
std = float_data[:200000].std(axis=0)
float_data /= std

There is a generator function for creating the dataset (I've read tensorflow.keras.utils.Sequence is a preferred choice but I failed in converting this generator to Sequence subclass

def generator(data, lookback, delay, min_index, max_index, shuffle=False, batch_size=128, step=6):
  if max_index is None:
    max_index = len(data) - delay - 1
  i = min_index + lookback

  while 1:
    if shuffle:
      rows = np.random.randint(min_index + lookback, max_index, size=batch_size)
    else:
      if i + batch_size >= max_index:
        i = min_index + lookback

      rows = np.arange(i, min(i + batch_size, max_index))
      i += len(rows)

    samples = np.zeros((len(rows),lookback // step,data.shape[-1]))
    targets = np.zeros((len(rows),))

    for j, row in enumerate(rows):
      indices = range(rows[j] - lookback, rows[j], step)
      samples[j] = data[indices]
      targets[j] = data[rows[j] + delay][1]
    yield samples, targets

Here are the parameter details

1) data —The original array of floating-point data, normalized

2) lookback —How many timesteps back the input data should go.

3) delay —How many timesteps in the future the target should be.

4) min_index and max_index —Indices in the data array that delimit which timesteps to draw from. This is useful for keeping a segment of the data for validation and another for testing.

5) shuffle —Whether to shuffle the samples or draw them in chronological order.

6) batch_size —The number of samples per batch.

7) step —The period, in timesteps, at which you sample data. You'll set it to 6 in order to draw one data point every hour.

And these generators

lookback = 1440
step = 6
delay = 144
batch_size = 128

train_gen = generator(float_data,
lookback=lookback,
delay=delay,
min_index=0,
max_index=200000,
shuffle=True,
step=step,
batch_size=batch_size)

val_gen = generator(float_data,
lookback=lookback,
delay=delay,
min_index=200001,
max_index=300000,
step=step,
batch_size=batch_size)

test_gen = generator(float_data,
lookback=lookback,
delay=delay,
min_index=300001,
max_index=None,
step=step,
batch_size=batch_size)

val_steps = (300000 - 200001 - lookback)
test_steps = (len(float_data) - 300001 - lookback)

The network layout is as follows

model = Sequential()
model.add(layers.Flatten(input_shape=(lookback // step, float_data.shape[-1])))
model.add(layers.Dense(32, activation='relu'))
model.add(layers.Dense(1))
model.compile(optimizer=RMSprop(), loss='mae')
model.compile(optimizer=RMSprop(), loss='mae')
history = model.fit(train_gen,steps_per_epoch=500,epochs=20,validation_data=test_gen,validation_steps=test_steps)

However the model gets stuck at

Train for 500 steps, validate for 119110 steps
Epoch 1/20
497/500 [============================>.] - ETA: 0s - loss: 0.3524

My model gets stuck while evaluating on the validation set. To make sure train_gen and val_gen where okay I tried

next(train_gen)
next(val_gen)

And they both display different values each time like

(array([[[ 0.34593055,  0.49507501,  0.4628141 , ...,  0.16203687,
           0.18470667,  0.84378526],
         [ 0.36243914,  0.6283707 ,  0.59460993, ...,  0.2921889 ,
           0.94414397,  0.60710086],
         [ 0.35182647,  0.64305582,  0.60912981, ...,  1.78242962,
           1.59631612,  0.43507171],
         ...,

What is wrong here?

Answer 1

The number of validation steps looks suspicious, as it is not being calculated with the batch size, so it is larger than it should be which would extend the validation phase in time considerably. The solution is to divide the number of steps with the batch size:

val_steps = val_steps // batch_size
test_steps = test_steps // batch_size

This will make steps have the right value.