Menu
I'm trying to predict price by characteristics. I chose a pretty simple model, but it works very strange. Loss function is extremely high and I can't see where the problem is.
Here is my model:
# define base model
def baseline_model():
# create model
model = Sequential()
model.add(Dense(62, input_dim = 62, kernel_initializer='normal', activation='relu'))
model.add(Dense(31, kernel_initializer='normal', activation='relu'))
model.add(Dense(15, kernel_initializer='normal', activation='relu'))
model.add(Dense(1, kernel_initializer='normal'))
# Compile model
model.compile(loss='mean_squared_error', optimizer='adam')
return model
That's how I prepare the data: (One-Hot and I split all data to train and test)
df = encode_onehot(dataframe, cols=['Shape', 'Cut', 'Color', 'Clarity', 'Polish', 'Symmetry', 'Culet', '\tFluorescence'])
dataset = df.values
X = dataset[1:,4:66]
Y = dataset[1:,2]
X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.25, random_state=42)
Finally, training:
baseline_model().fit(X_train, y_train, epochs=10, batch_size=64)
scores = baseline_model().evaluate(X_test, y_test, verbose=0)
print(baseline_model().summary())
And results are very sad:
Epoch 1/10
149767/149767 [==============================] - 4s - loss: 104759338.0333
Epoch 2/10
149767/149767 [==============================] - 4s - loss: 104594236.9627
Epoch 3/10
149767/149767 [==============================] - 4s - loss: 104556662.2948
And it doesn't get better.
What am I doing wrong?
999
Loss is a value that represents the summation of errors in our model. It measures how well (or bad) our model is doing. If the errors are high, the loss will be high, which means that the model does not do a good job. Otherwise, the lower it is, the better our model works.
Typically the validation loss is greater than training one, but only because you minimize the loss function on training data. I recommend to use something like the early-stopping method to prevent the overfitting. The results of the network during training are always better than during verification.
categorical_crossentropy: Used as a loss function for multi-class classification model where there are two or more output labels. The output label is assigned one-hot category encoding value in form of 0s and 1. The output label, if present in integer form, is converted into categorical encoding using keras.
Most of the time we would observe that accuracy increases with the decrease in loss -- but this is not always the case. Accuracy and loss have different definitions and measure different things. They often appear to be inversely proportional but there is no mathematical relationship between these two metrics.
As @Yu-Yang said you are using mean squared error as loss function. I had this same problem before where the loss value will be very large, on changing the loss function to mean_squared_logarithmic_error, I got the desired result.
model %>% compile(
optimizer = optimizer_rmsprop(lr=0.0001),
loss = loss_mean_squared_logarithmic_error,
metrics = c("accuracy")
)
The loss value changed to
Epoch 1/10
326981/326981 [==============================] - 17s - loss: 0.0048 - acc: 0.9896
Hope this finds useful !
91
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With
