Keras, simple CNN with binary classification doesn't converge — minimal example (from DeepLizard)

Question

I am new to NN and Keras. I have tried this tutorial from DeepLizard and followed it until the lady showed the first training results ~30min in. I followed the steps exactly except the lines with 'shuffle' since that was not how it is supposed to be used (to my knowledge). Here is my code:

import numpy as np
import random as random
from random import randint
from sklearn.preprocessing import MinMaxScaler

from random import shuffle

train_labels = []
train_samples = []

for i in range(50):
  random_younger = randint(13,64)
  train_samples.append(random_younger)
  train_labels.append(1)

  random_older = randint(65,100)
  train_samples.append(random_older)
  train_labels.append(0)

for i in range (1000):
  random_younger = randint(13,64)
  train_samples.append(random_younger)
  train_labels.append(0)

  random_older = randint(65,100)
  train_samples.append(random_older)
  train_labels.append(1)

train_samples = np.array(train_samples)
train_labels = np.array(train_labels)
shuffle(train_samples)
shuffle(train_labels)

scaler = MinMaxScaler(feature_range = (0,1))
scaled_train_samples = scaler.fit_transform(train_samples. reshape(-1,1))

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Activation, Dense
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.metrics import categorical_crossentropy

model = Sequential([Dense(units=16, input_shape=(1,), activation='relu'),
                Dense(units=32, activation='relu'),
                Dense(units=2, activation='softmax')])

model.summary()

    model.compile(optimizer=Adam(learning_rate=0.0001),loss='sparse_categorical_crossentropy', metrics=['accuracy'])

model.fit(x=scaled_train_samples, y=train_labels, batch_size=5, epochs=30, shuffle=True, verbose=2)

My NN doesn't learn staying at 50% accuracy all the time. I tried to vary the #of layers and epochs, I have checked the data types and structure. Can someone help with the first in my life minimal example? Thanks!

Try representing the ages in binary, arrays of zeros and ones, instead than using integers — alec_djinn, Mar 31 '21 at 18:04
I actually think shuffle is not needed there at all. I think since "model.fit" shuffles by default, it was a bug in her code. The code, as she wrote it, doesn't work, but if I delete the lines with "shuffle" everything works as expected. — MsTais, Mar 31 '21 at 18:09
@NicolasGervais is right. This part needs to be modified: shuffle(train_samples) shuffle(train_labels) This thread could help for your case:https://stackoverflow.com/questions/23289547/shuffle-two-list-at-once-with-same-order — Dr. H. Lecter, Mar 31 '21 at 18:36

score 1 · Accepted Answer · answered Mar 31 '21 at 18:34

As mentioned before, I was shuffling x and y independently. In the tutorial I have reference "shuffle" is used the way it is not supposed to work, see the documentation. My solution:

c = list(zip(train_samples, train_labels))

shuffle(c)

train_samples, train_labels = zip(*c)
train_samples = np.array(train_samples)
train_labels=np.array(train_labels)

Keras, simple CNN with binary classification doesn't converge — minimal example (from DeepLizard)

1 Answers1