1

This is a long question because i try to explain the more i can my problem because it's a recurrent problem for me and i really don't understand so thank you for taking the time to read me

I want to create a sequential dense model which takes as input list with dimension like this:

[batch_size, data_dimension]

So i defined my network like this:

ModelDense = Sequential()

ModelDense.add(Dense(380, input_shape=(None,185), activation='elu', kernel_initializer='glorot_normal'))
ModelDense.add(Dense(380, activation='elu', kernel_initializer='glorot_normal'))
ModelDense.add(Dense(380, activation='elu', kernel_initializer='glorot_normal'))
ModelDense.add(Dense(7, activation='elu', kernel_initializer='glorot_normal'))
optimizer = tf.keras.optimizers.Adam(lr=0.00025)

ModelDense.compile(loss='mean_squared_error', optimizer=optimizer, metrics=['accuracy'])

but when i use this network with input shaped like this : (1, 185) i got the error:

Error when checking input: expected dense_input to have 3 dimensions, but got array with shape (185, 1)

Don't ask me why i said that my vector shape is (1, 185) and in the error message we see (185, 1) because when i check my array shape just before giving it as input to my network the shape shown is (1, 185)

Ok, so i checked some topics then i found this one in which is explained that :

Dense layers require inputs as (batch_size, input_size) or (batch_size, optional,...,optional, input_size)

So that is what i did idn't it ? But i also saw that:

Shapes in Keras :

...

So, even if you used input_shape=(50,50,3), when keras sends you messages, or when you print the model summary, it will show (None,50,50,3)

...

So, when defining the input shape, you ignore the batch size: input_shape=(50,50,3)

Ok ! let's try i now definied my input layer like this :

ModelDense.add(Dense(380, input_shape=(185,), activation='elu', kernel_initializer='glorot_normal'))

When i do a model.summary() :

_________________________________________________________________ Layer (type) Output Shape Param # ================================================================= dense (Dense) (None, 380) 70680 _________________________________________________________________ dense_1 (Dense) (None, 380) 144780 _________________________________________________________________ dense_2 (Dense) (None, 380) 144780 _________________________________________________________________ dense_3 (Dense) (None, 7) 2667 ================================================================= Total params: 362,907 Trainable params: 362,907 Non-trainable params: 0

Ok i think it's that i want but when i give THE SAME array as input i now get the error:

ValueError: Error when checking input: expected dense_input to have shape (185,) but got array with shape (1,)

I'm confused, what am i misunderstanding ?

_________EDIT__________ :

Prediction function:

def predict(dense_model, state, action_size, epsilon):

    alea = np.random.rand()

    # DEBUG
    print(state)
    print(np.array(state).shape)

    output = dense_model.predict(state)

    if (epsilon > alea):
        action = random.randint(1, action_size) - 1
        flag_alea = True

    else:
        action = np.argmax(output)
        flag_alea = False

    return output, action, flag_alea

The line where i use my function:

Qs, action, flag_alea = predict(Dense_model, [state], ACTION_SIZE, Epsilon)

Exact result of my 'DEBUG' printing:

[[0.0, 0.0, 0.0, 0.12410027302060064, 0.0, 0.0, 0.0, 0.0, 0.0, 0.18851780241253108, 0.0, 0.0, 0.2863141820958198, 0.0, 0.07328154770628756, 0.418848167539267, 0.07328154770628756, 0.2094240837696335, 0.42857142857142855, 0.0, 0.12410027302060064, 0.0, 0.0, 0.0, 0.0, 0.263306220774655, 0.14740566037735847, 0.40346984062941293, 0.675310642895732, 0.0, 0.0, 0.0, 0.0, 0.07328154770628756, 0.0, 0.4396892862377253, 0.0, 0.42857142857142855, 0.0, 0.12410027302060064, 0.08759635599159075, 0.0, 0.1401927621025243, 0.6755559204272007, 0.0, 0.0, 0.11564568886156315, 0.4051863857374392, 0.0, 0.0, 0.19087612139721322, 0.0, 0.07328154770628756, 0.6282722513089005, 0.14656309541257512, 0.10471204188481675, 0.42857142857142855, 0.0, 0.12410027302060064, 0.0, 0.0, 0.0, 0.0, 0.0974621385076755, 0.0, 0.0, 0.675310642895732, 0.0, 0.0, 0.0, 0.09543806069860661, 0.07328154770628756, 0.10471204188481675, 0.5129708339440129, 0.5233396901920598, 0.42857142857142855, 0.0, 0.0, 0.0, 0.0, 0.5528187746700128, 0.6755564266434103, 0.0, 0.0, 0.10086746015735323, 0.1350621285791464, 0.0, 0.0, 0.0, 0.0, 0.14891426591693724, 0.5166404112353377, 0.14656309541257512, 0.10471204188481675, 0.42857142857142855, 0.00846344605088234, 0.012550643645226955, 0.0, 0.0, 0.004527776502072811, 0.0, 0.001294999849051237, 0.019391579553484917, 0.02999694086611271, 0.0026073455810546875, 0.0, 0.0, 0.016546493396162987, 0.024497902020812035, 0.00018889713101089, 0.0, 0.005568447522819042, 0.0, 0.007975691929459572, 0.01434263214468956, 0.0, 6.733229383826256e-05, 0.0012099052546545863, 0.0, 0.0001209513284265995, 0.01868056133389473, 0.025530844926834106, 0.004079729784280062, 0.0, 0.0, 0.01332627609372139, 0.026645798236131668, 0.0, 0.0, 0.007684763520956039, 0.0, 0.010554256848990917, 0.007236589677631855, 0.0013368092477321625, 0.000697580398991704, 0.00213554291985929, 0.0, 0.0021772112231701612, 0.012761476449668407, 0.015171871520578861, 0.001512336079031229, 0.0, 0.0, 0.008273545652627945, 0.01777557097375393, 0.006600575987249613, 0.0, 0.007174563594162464, 0.0, 0.004660750739276409, 0.009024208411574364, 0.0, 0.0014235835988074541, 0.0, 0.0, 0.0, 0.008785379119217396, 0.010602384805679321, 0.0024691042490303516, 0.0, 0.0, 0.003091508522629738, 0.0120345214381814, 0.003123666625469923, 0.0, 0.005664713680744171, 0.0, 0.004825159907341003, 0.0034197410568594933, 0.0030767947901040316, 0.004110954236239195, 0.0, 0.0, 0.001896441332064569, 0.002400417113676667, 0.0012791997287422419, 0.0, 0.0, 0.0, 0.0021027529146522284, 0.006922871805727482, 0.004868669901043177, 0.0, 7.310241926461458e-05, 0.0]]

(1, 185)

_________EDIT2__________ :

Error traceback:

File ".!Qltrain.py", line 360, in Qs, action, flag_alea = predict(Dense_model, [state], ACTION_SIZE, Epsilon) File ".\Lib\Core.py", line 336, in predict output = dense_model.predict(state) File "C:\Users\Odeven\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1096, in predict x, check_steps=True, steps_name='steps', steps=steps) File "C:\Users\Odeven\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 2382, in _standardize_user_data exception_prefix='input') File "C:\Users\Odeven\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training_utils.py", line 362, in standardize_input_data ' but got array with shape ' + str(data_shape)) ValueError: Error when checking input: expected dense_input to have shape (185,) but got array with shape (1,)

If u check out the first 3 lines you can see that the code from where the erorr is coming is the code i added in my first edit

_______self-containing example_______

Content of test.py:

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
import random
import numpy as np

ModelDense = Sequential()

ModelDense.add(Dense(380, input_shape=(185,), activation='elu', kernel_initializer='glorot_normal'))
ModelDense.add(Dense(380, activation='elu', kernel_initializer='glorot_normal'))
ModelDense.add(Dense(380, activation='elu', kernel_initializer='glorot_normal'))
ModelDense.add(Dense(7, activation='elu', kernel_initializer='glorot_normal'))
optimizer = tf.keras.optimizers.Adam(lr=0.00025)

ModelDense.compile(loss='mean_squared_error', optimizer=optimizer, metrics=['accuracy'])


ModelDense.summary()



def predict(dense_model, state, action_size, epsilon):

    alea = np.random.rand()

    print(state)
    print(np.array(state).shape)

    dense_model.summary()

    output = dense_model.predict(state)

    if (epsilon > alea):
        action = random.randint(1, action_size) - 1
        flag_alea = True

    else:
        action = np.argmax(output)
        flag_alea = False

    return output, action, flag_alea



state = []
state.append([np.random.rand()] * 185)
output, ac, flag = predict(ModelDense, state, 7, 0.0)

print(output)

Complete output:

_________________________________________________________________ Layer (type) Output Shape Param # ================================================================= dense (Dense) (None, 380) 70680 _________________________________________________________________ dense_1 (Dense) (None, 380) 144780 _________________________________________________________________ dense_2 (Dense) (None, 380) 144780 _________________________________________________________________ dense_3 (Dense) (None, 7) 2667 ================================================================= Total params: 362,907 Trainable params: 362,907 Non-trainable params: 0 _________________________________________________________________ [[0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739, 0.11966889292971739]] (1, 185) _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= dense (Dense) (None, 380) 70680 _________________________________________________________________ dense_1 (Dense) (None, 380) 144780 _________________________________________________________________ dense_2 (Dense) (None, 380) 144780 _________________________________________________________________ dense_3 (Dense) (None, 7) 2667 ================================================================= Total params: 362,907 Trainable params: 362,907 Non-trainable params: 0 _________________________________________________________________ Traceback (most recent call last): File ".\test.py", line 47, in output, ac, flag = predict(ModelDense, state, 7, 0.0) File ".\test.py", line 31, in predict output = dense_model.predict(state) File "C:\Users\Odeven\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1096, in predict x, check_steps=True, steps_name='steps', steps=steps) File "C:\Users\Odeven\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py", line 2382, in _standardize_user_data exception_prefix='input') File "C:\Users\Odeven\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\keras\engine\training_utils.py", line 362, in standardize_input_data ' but got array with shape ' + str(data_shape)) ValueError: Error when checking input: expected dense_input to have shape (185,) but got array with shape (1,)

Xeyes
  • 583
  • 5
  • 25
  • Are you sure that your training input has shape (1, 158)? Both errors point that the actual shape is (158, 1) – Dr. Snoopy Jul 10 '19 at 08:37
  • I add my prediction function to my post, i also add the line where i'm using it and the debug message i have when i print the dimension – Xeyes Jul 10 '19 at 08:42
  • Are you sure that this is the part of the code that produces the error? Maybe you have an (185,1) array somewhere else and it is error'ing there. For me its not clear how the error and the code you included are related. – Dr. Snoopy Jul 10 '19 at 08:58
  • I am completely sure about the code executed, i add the traceback of the error, i know it's really weird that is why i really don't uderstand this error – Xeyes Jul 10 '19 at 09:02
  • Then make a self-containing example that reproduces the error and we can run. – Dr. Snoopy Jul 10 '19 at 09:11
  • Ok i just created a file named test.py in which i only wrote the important code and i get the same error, i add the content of this file in a third edit + the error traceback – Xeyes Jul 10 '19 at 09:18
  • Let us [continue this discussion in chat](https://chat.stackoverflow.com/rooms/196249/discussion-between-xeyes-and-matias-valdenegro). – Xeyes Jul 10 '19 at 09:33

2 Answers2

2

Change this:

output = dense_model.predict(state)

Into this:

output = dense_model.predict(np.array(state))

Seems keras gets confused if you pass a plain list to predict and might not do what you want, this way you make sure that state is a numpy array of the shape you expect.

Dr. Snoopy
  • 55,122
  • 7
  • 121
  • 140
  • Ok thank a lot ! that worked well ! I find very suprising that keras isn't interpreting a list correctly but ok i'll now always convert my list ^^. Thank for taking the time to read such a long post – Xeyes Jul 10 '19 at 11:32
0

I am not a pro of Keras but I think, as @Matias Valdenegro said and since the algorithm is going to perform matrix multiplications through the network, it is expecting you to give an array.

If you have only one state on which you want to perform prediction you can expand the dim of your data as follow : 

state = np.expand_dims(state, axis=0)
Adakor
  • 65
  • 6
  • When i do what you explained on a 'non numpy array' container i got the following erorr : ValueError: Error when checking input: expected dense_input to have 2 dimensions, but got array with shape (1, 1, 185)... so i think the best solution just is to convert my list into array :) – Xeyes Jul 10 '19 at 11:38