After calling model.fit(), the loss is NaN (checked for NaN values after reading data with pandas)

TensorFlow General Discussion
1

I have not yet figured out why, but every time I fit the model, the loss is shown to be NaN, though I have thoroughly (tried) to drop any NaN rows in the data that I am training against. This is my first time trying to learn ML, so patience with me would be appreciated.

I will walk through my code step by step here in an effort to better explain my choices:

  1. Import dependencies:
import pandas as pd
from sklearn.model_selection import train_test_split
from keras.models import Sequential
from keras.layers import Dense
from sklearn.metrics import accuracy_score
from sklearn.linear_model import Ridge
from sklearn.preprocessing import StandardScaler
from sklearn.impute import SimpleImputer
  1. Read data, drop any rows containing NaN data, then check again for NaN rows
df = pd.read_csv('Wednesday-workingHours.pcap_ISCX.csv')
df = df.dropna()
df.drop([' Label'], axis=1).isna().sum()
  1. Split data & interpret target data
X = pd.get_dummies(df.drop([' Label'], axis=1))
Y = df[' Label'].apply(lambda x: 1 if x=='BENIGN' else 0)
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)
  1. Set the type of model & add layers
model = Sequential()
model.add(Dense(units=32, activation='relu', input_dim=78))
model.add(Dense(units=64, activation='relu'))
model.add(Dense(units=1, activation='sigmoid'))
  1. Compile the model
model.compile(loss='binary_crossentropy', optimizer="adam", metrics='accuracy')
  1. Fit the model
model.fit(X_train, Y_train, epochs=200, batch_size=100)

For reference, the dataset I am using for this is can be found here which has 692,704 rows and 79 columns.

Sample output of step 6:

Epoch 1/100
554/554 [==============================] - 1s 1ms/step - loss: nan - accuracy: 0.3652

As you could probably guess from the list of imports, I have tried using a scalar, imputer, and L2 Regularization to correct this problem, but it still persists and I have run out of ideas.

My attempts at these corrections were removed after each fail, but interestingly enough when I switch to L2 Reg. I get a value error: ValueError: Input X contains infinity or a value too large for dtype('float64').

I also took a look at my model.summary() and didn’t find anything too concerning:

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 dense_3 (Dense)             (None, 32)                2528      
                                                                 
 dense_4 (Dense)             (None, 64)                2112      
                                                                 
 dense_5 (Dense)             (None, 1)                 65        
                                                                 
=================================================================
Total params: 4705 (18.38 KB)
Trainable params: 4705 (18.38 KB)
Non-trainable params: 0 (0.00 Byte)

Again this is my first time with ML and I’m basing this code off of reading and tutorials.

3

Hi @Christopher_Calvani

Welcome to the TensorFlow Forum!

Please share the ‘Wednesday-workingHours.pcap_ISCX.csv’ file (if it is shareable) to check and understand the dataset entities format which could be the reason of computational error in model training and for the ‘NAN’ metrics values.