I am trying to train a Conv1D network, and this is my code:
# %% Imports
import time as time
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
from keras.callbacks import ModelCheckpoint, EarlyStopping
from keras.layers import Conv1D
from keras.layers import Dense
from keras.layers import Flatten
from keras.layers import InputLayer
from keras.layers import MaxPooling1D
from tensorflow.keras import Model
from keras.models import Sequential
from keras import backend as K
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelBinarizer
from sklearn.ensemble import RandomForestClassifier
# %% Defining classification metrics (precision, recall and f1_score)
def recall_m(y_true_m, y_pred_m):
true_positives = K.sum(K.round(K.clip(y_true_m * y_pred_m, 0, 1)))
possible_positives = K.sum(K.round(K.clip(y_true_m, 0, 1)))
recall_metrica = true_positives / (possible_positives + K.epsilon())
return recall_metrica
def precision_m(y_true_m, y_pred_m):
true_positives = K.sum(K.round(K.clip(y_true_m * y_pred_m, 0, 1)))
predicted_positives = K.sum(K.round(K.clip(y_pred_m, 0, 1)))
precision_metrica = true_positives / (predicted_positives + K.epsilon())
return precision_metrica
def f1_m(y_true_m, y_pred_m):
precision_met = precision_m(y_true_m, y_pred_m)
recall_met = recall_m(y_true_m, y_pred_m)
return 2*((precision_met*recall_met)/(precision_met+recall_met+K.epsilon()))
# %% Loading dataset
# percentuais_amostras = [0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08,
# 0.09, 0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 1]
percentuais_amostras = [1]
rede = 'Conv1D'
rede_rf = 'Conv1D_RF'
dataset = 'M7'
modelo = 'M9'
for perc_amostras in percentuais_amostras:
df = pd.read_csv(f'C:/Users/ldsp_/OneDrive/Área de Trabalho/Irish/Dataset_{dataset}/Dataset_{dataset}_{perc_amostras}.txt', sep=' ')
saida = df['FLAG']
# %% Split into inputs and outputs
x = df.iloc[:, :-1].values
y = df['FLAG'].values
del df
# %% Encoding outputs
label_bin = LabelBinarizer()
classes_unicas = np.unique(y)
label_bin.fit_transform(classes_unicas)
# %% Train/test split
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=42)
del x, y
# One-hot-encoding output
y_train_one_hot = label_bin.transform(y_train)
# Reshaping input to Conv1D network format
x_train = x_train.reshape(x_train.shape[0], x_train.shape[1], 1)
# %% Conv1D
feature_extractor = Sequential(name=f'{rede}_{modelo}_{perc_amostras}')
feature_extractor.add(InputLayer((x_train.shape[1], 1), name='Camada_Entrada'))
feature_extractor.add(Conv1D(filters=64, kernel_size=2, activation='relu', name='1_Camada_Conv'))
feature_extractor.add(MaxPooling1D(pool_size=2))
feature_extractor.add(Conv1D(filters=128, kernel_size=2, activation='relu', name='2_Camada_Conv'))
feature_extractor.add(MaxPooling1D(pool_size=3))
feature_extractor.add(Conv1D(filters=64, kernel_size=2, activation='tanh', name='3_Camada_Conv'))
feature_extractor.add(MaxPooling1D(pool_size=3))
feature_extractor.add(Flatten())
fe_output = feature_extractor.output # Goes to RF_Classifier
dense_1 = Dense(units=4 * y_train_one_hot.shape[1], activation='relu', name='1_Dense')(fe_output)
output_layer = Dense(units=y_train_one_hot.shape[1], activation='softmax', name='Camada_Saida')(dense_1)
# Compile
cnn_model = Model(inputs=feature_extractor.input, outputs=output_layer)
cnn_model.compile(optimizer='adam', loss='categorical_crossentropy',
metrics=['accuracy',precision_m, recall_m, f1_m])
cnn_model.summary()
# %% Checkpoint and Earlystop
cp = ModelCheckpoint(
filepath=f'C:/Users/ldsp_/OneDrive/Área de Trabalho/Irish/{rede}/Resultados {rede}_{modelo}/{rede}_{modelo}_{perc_amostras}.hdf5',
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min')
earlystop = EarlyStopping(monitor='val_loss',
patience=50,
verbose=1,
mode='min',
restore_best_weights=True)
# %% Treinando a CNN
val_size = 0.2
t = time.time()
history = cnn_model.fit(x_train,
y_train_one_hot,
batch_size=256,
epochs=250,
callbacks=[cp, earlystop],
validation_split=val_size)
tempo_modelo = round((time.time() - t) / 60, 4)
What i am trying to do here, is to train this Conv1D network, and use feature_extractor as input to a RandomForestClassifier from Sklearn.
But after the first epoch, i am getting this error:
Epoch 1/250
10372/10372 [==============================] - 40s 3ms/step - loss: 0.5776 - accuracy: 0.7931 - precision_m: 0.8696 - recall_m: 0.7234 - f1_m: 0.7868 - val_loss: 0.4846 - val_accuracy: 0.8222 - val_precision_m: 0.8831 - val_recall_m: 0.7699 - val_f1_m: 0.8225
Epoch 00001: val_loss improved from inf to 0.48459, saving model to C:/Users/ldsp_/OneDrive/Área de Trabalho/Irish/Conv1D/Resultados Conv1D_M9\Conv1D_M9_1.hdf5
Traceback (most recent call last):
File ~\miniconda3\envs\tf-gpu\lib\site-packages\spyder_kernels\py3compat.py:356 in compat_exec
exec(code, globals, locals)
File c:\users\ldsp_\onedrive\área de trabalho\irish\conv1d\conv1d_m9.py:189
history = cnn_model.fit(x_train,
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\engine\training.py:1229 in fit
callbacks.on_epoch_end(epoch, epoch_logs)
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\callbacks.py:435 in on_epoch_end
callback.on_epoch_end(epoch, logs)
File ~\miniconda3\envs\tf-gpu\lib\site-packages\keras\callbacks.py:1376 in on_epoch_end
self._save_model(epoch=epoch, logs=logs)
File ~\miniconda3\envs\tf-gpu\lib\site-packages\keras\callbacks.py:1428 in _save_model
self.model.save(filepath, overwrite=True, options=self._options)
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\engine\training.py:2111 in save
save.save_model(self, filepath, overwrite, include_optimizer, save_format,
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\saving\save.py:146 in save_model
hdf5_format.save_model_to_hdf5(
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\saving\hdf5_format.py:110 in save_model_to_hdf5
model_metadata = saving_utils.model_metadata(model, include_optimizer)
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\saving\saving_utils.py:150 in model_metadata
model_config['config'] = model.get_config()
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\engine\functional.py:649 in get_config
return copy.deepcopy(get_network_config(self))
File ~\miniconda3\envs\tf-gpu\lib\site-packages\tensorflow\python\keras\engine\functional.py:1368 in get_network_config
new_node_index = node_conversion_map[node_key]
KeyError: 'Camada_Entrada_ib-0'
One thing i have noticed is that i get this from model.summary():
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
Total params: 36,860
Trainable params: 36,860
Non-trainable params: 0
_________________________________________________________________
The parameteres are there, but no output shape?
I have tried this same code in another PC, and it works. What could be the reason i am getting this error in this case?