Keras 回归模型建议

Question

I am trying to solve a regression with Keras but MSE is huge, I mean like 29346217.6819

I am really new, so do you have any suggestions to make the model give reasonable mse? I am not sure even my data is OK or problematic but those are actual sales data.

Data (about to 3000 lines. I use 2000 for training and 1000 for testing) Full data is here

ProductNo,Day,Month,CartonSales
1,6,02,2374
1,3,02,2374
1,6,04,2374
1,6,04,2374
1,3,06,2374
1,6,09,2374
1,1,09,2374
1,6,09,2374
1,6,10,2374

Code

from keras import optimizers
from keras.callbacks import Callback
from numpy import array
from keras.models import Sequential
from keras.layers import Dense, Dropout
from matplotlib import pyplot
import pandas as pds
# prepare sequence


class TestCallback(Callback):
    def __init__(self, test_data):
        self.test_data = test_data

    def on_epoch_end(self, epoch, logs={}):
        x, y = self.test_data
        loss, acc = self.model.evaluate(x, y, verbose=0)
        print('\nTesting loss: {}, acc: {}\n'.format(loss, acc))

dataframe = pds.read_csv('pmidata.csv', usecols=[0, 1, 2, 3])
dataframe = dataframe.sample(frac=1)

dataframeX_train = dataframe.iloc[0:2000][['ProductNo', 'Day', 'Month']]
dataframeY_train = dataframe.iloc[0:2000][['CartonSales']]

dataframeX_test = dataframe.iloc[2001:3001][['ProductNo', 'Day', 'Month']]
dataframeY_test = dataframe.iloc[2001:3001][['CartonSales']]

# create model
model = Sequential()
model.add(Dense(3, input_dim=3, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(1))
model.compile(loss='mse', optimizer='adam', metrics=['mse'])
#sgd = optimizers.SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
#model.compile(loss='mse', optimizer=sgd, metrics=['mse'])
# train model
#history = model.fit(dataframe, dataframe, epochs=500, batch_size=len(X), verbose=2)
history = model.fit(dataframeX_train, dataframeY_train, epochs=100, batch_size=4, verbose=2, callbacks=[TestCallback((dataframeX_test, dataframeY_test))])
# plot metrics
pyplot.plot(history.history['mean_squared_error'])
pyplot.show()

Answer 1

As far as i can tell from your code above, your y values are CartonSales. Sales can have large values and large range and that's probably why you get such a high error. You could use mean_squared_logarithmic_error instead of mean square error but i would suggest to do the following.

Continue using mean square error. log transform you y values and later exp transform you predictions

import numpy as np
dataframeY_train = np.log(dataframeY_train)
dataframeY_test = np.log(dataframeY_test )
....
predictions=model.predict(dataframeX_test)[:,0]
predictions = np.exp(predictions)