Keras and the input layer - machine-learning

Question

So I'm trying to learn ANN's with Keras as I heard it is simpler that Theano or TensorFlow. I have a number of questions the first is to do with the input layer.
So far I have this line of code as the input:
model.add(Dense(3 ,input_shape=(2,), batch_size=50 ,activation='relu'))
Now the data I want to add into the model is of the following shape:
Index(['stock_price', 'stock_volume', 'sentiment'], dtype='object')
[[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 1.42857143e-01]
[ 3.01440000e+02 7.87830000e+04 5.88235294e-02]
[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 0.00000000e+00]
[ 3.01440000e+02 7.87830000e+04 5.26315789e-02]]
I want to make a model see if I can find a correlation between stock prices and tweet sentiment and I just threw volume in there because eventually, I want to see if it can find a pattern with that as well.
So my second question is after running my input layer with several different parameters I get this problem which I can't explain. So when I run this line:
model.add(Dense(3 ,input_shape=(2,), batch_size=50 ,activation='relu'))
with the following line I get this output error:
ValueError: Error when checking model input: expected dense_1_input to have shape (50, 2) but got array with shape (50, 3)
But when I change the input shape to the requested '3' I get this error:
ValueError: Error when checking model target: expected dense_2 to have shape (50, 1) but got array with shape (50, 302)
Why has the 2 changed into '302' on the error message?
I'm probably overlooking some really basic problems since this is the first neural net I've tried to implement because I've only used the application for of Weka before.
Anyway here is a copy of my full code:
from keras.models import Sequential, Model
from keras.layers import Dense, Activation, Input
from keras.optimizers import SGD
from keras.utils import np_utils
import pymysql as mysql
import pandas as pd
import config
import numpy
import pprint
model = Sequential()
try:
sql = "SELECT stock_price, stock_volume, sentiment FROM tweets LIMIT 50"
con = mysql.connect(config.dbhost, config.dbuser, config.dbpassword, config.dbname, charset='utf8mb4', autocommit=True)
results = pd.read_sql(sql=sql, con=con, columns=['stock_price', 'stock_volume', 'sentiment'])
finally:
con.close()
npResults = results.as_matrix()
cols = np_utils.to_categorical(results['stock_price'].values)
data = results.values
print(cols)
# inputs:
# 1st = stock price
# 2nd = tweet sentiment
# 3rd = volume
model.add(Dense(3 ,input_shape=(3,), batch_size=50 ,activation='relu'))
model.add(Dense(20, activation='linear'))
sgd = SGD(lr=0.3, decay=0.01, momentum=0.2)
model.compile(loss='sparse_categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
model.summary()
model.fit(x=data, y=cols, epochs=100, batch_size=100, verbose=2)
EDIT:
Here is all the output I get fom the console:
C:\Users\Def\Anaconda3\python.exe C:/Users/Def/Dropbox/Dissertation/ann.py
Using Theano backend.
C:\Users\Def\Dropbox\Dissertation
[[ 0. 0. 0. ..., 0. 0. 1.]
[ 0. 0. 0. ..., 0. 0. 1.]
[ 0. 0. 0. ..., 0. 0. 1.]
...,
[ 0. 0. 0. ..., 0. 0. 1.]
[ 0. 0. 0. ..., 0. 0. 1.]
[ 0. 0. 0. ..., 0. 0. 1.]]
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense_1 (Dense) (50, 3) 12
_________________________________________________________________
dense_2 (Dense) (50, 20) 80
=================================================================
Traceback (most recent call last):
File "C:/Users/Def/Dropbox/Dissertation/ann.py", line 38, in <module>
model.fit(x=data, y=cols, epochs=100, batch_size=100, verbose=2)
File "C:\Users\Def\Anaconda3\lib\site-packages\keras\models.py", line 845, in fit
initial_epoch=initial_epoch)
File "C:\Users\Def\Anaconda3\lib\site-packages\keras\engine\training.py", line 1405, in fit
batch_size=batch_size)
File "C:\Users\Def\Anaconda3\lib\site-packages\keras\engine\training.py", line 1299, in _standardize_user_data
exception_prefix='model target')
File "C:\Users\Def\Anaconda3\lib\site-packages\keras\engine\training.py", line 133, in _standardize_input_data
str(array.shape))
ValueError: Error when checking model target: expected dense_2 to have shape (50, 20) but got array with shape (50, 302)
Total params: 92.0
Trainable params: 92
Non-trainable params: 0.0
_________________________________________________________________
Process finished with exit code 1

Answer 1

I think you are using the wrong metric: sparse_categorical_crossentropy
Is there a reason you prefer this over the normal: categorical_crossentropy ?
When using categorical_crossentropy, you should encode your targets in 1-hot coding fasion (using for instance: cols = np_utils.to_categorical(results['stock_price'].values)).
On the other hand, sparse_categorical_crossentropy uses integer-based labels.
So either use:
cols = np_utils.to_categorical(results['stock_price'].values)
with
model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
and an output layer of (num-categories) neurons
or use:
cols = results['stock_price'].values.astype(np.int32)
with
model.compile(loss='sparse_categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
and an single-neuron output layer.