Ufdjrw

What do *foreign films* mean for an American?

Was it really inappropriate to write a pull request for the company I interviewed with?

Specifying a starting column with colortbl package and xcolor

When a wind turbine does not produce enough electricity how does the power company compensate for the loss?

Why does cron require MTA for logging?

What are some noteworthy "mic-drop" moments in math?

Are small insurances worth it?

What is Tony Stark injecting into himself in Iron Man 3?

When Schnorr signatures are part of Bitcoin will it be possible validate each block with only one signature validation?

What do you call someone who likes to pick fights?

Vocabulary for giving just numbers, not a full answer

Shifting between bemols and diesis in the key signature

Which classes are needed to have access to every spell in the PHB?

School performs periodic password audits. Is my password compromised?

What problems would a superhuman have who's skin is constantly hot?

Recommendation letter by significant other if you worked with them professionally?

How exactly does an Ethernet collision happen in the cable, since nodes use different circuits for Tx and Rx?

Trig Subsitution When There's No Square Root

Why is a very small peak with larger m/z not considered to be the molecular ion?

What is better: yes / no radio, or simple checkbox?

Why does Solve lock up when trying to solve the quadratic equation with large integers?

Expressing logarithmic equations without logs

In the late 1940’s to early 1950’s what technology was available that could melt a LOT of ice?

Having the player face themselves after the mid-game

Conv1D model for time series

0

I am a novice in the area of Deep Learning and am willing to build a Conv1D autoencoder for time series with such shapes:

• samples: 200


• timesteps: 23


• features: 178

I am not sure about how should I set the parameters: filters, kernel_size and the layers MaxPooling1D(what's the behaviour?), Flatten(what's the behaviour?), and how should be architecturally designed the simplest Conv1D Autoencoder.

I tried to gather something alone but my version is not working at all:

input_layer = Input(shape=(TIMESTEPS, feature_size))
# ENCODER
x = Conv1D(filters=encoding_dim, kernel_size=TIMESTEPS, activation='relu', padding='valid')(input_layer)
x1 = MaxPooling1D(poolsize=TIMESTEPS)(x)
flat = Flatten()(x1)
encoded = Dense(units=encoding_dim, activation = 'relu')(flat)

print("shape of encoded ".format(K.int_shape(encoded)))

# DECODER 
x_ = Conv1D(encoding_dim, TIMESTEPS, activation='relu', padding='valid')(encoded)
upsamp = UpSampling1D(TIMESTEPS)(x_)
flat = Flatten()(upsamp)
decoded = Dense(units=feature_size, activation = 'relu')(flat)
decoded = Reshape((TIMESTEPS, feature_size))(decoded)

print("shape of decoded ".format(K.int_shape(decoded)))

autoencoder = Model(input_layer, decoded)

ValueError: Negative dimension size caused by subtracting 23 from 1 for 'max_pooling1d/MaxPool' (op: 'MaxPool') with input shapes: [?,1,1,59].

python keras conv-neural-network autoencoder

share|improve this question

asked Mar 6 at 14:24

GuidoGuido

678

add a comment | 

0

I am a novice in the area of Deep Learning and am willing to build a Conv1D autoencoder for time series with such shapes:

• samples: 200


• timesteps: 23


• features: 178

I am not sure about how should I set the parameters: filters, kernel_size and the layers MaxPooling1D(what's the behaviour?), Flatten(what's the behaviour?), and how should be architecturally designed the simplest Conv1D Autoencoder.

I tried to gather something alone but my version is not working at all:

input_layer = Input(shape=(TIMESTEPS, feature_size))
# ENCODER
x = Conv1D(filters=encoding_dim, kernel_size=TIMESTEPS, activation='relu', padding='valid')(input_layer)
x1 = MaxPooling1D(poolsize=TIMESTEPS)(x)
flat = Flatten()(x1)
encoded = Dense(units=encoding_dim, activation = 'relu')(flat)

print("shape of encoded ".format(K.int_shape(encoded)))

# DECODER 
x_ = Conv1D(encoding_dim, TIMESTEPS, activation='relu', padding='valid')(encoded)
upsamp = UpSampling1D(TIMESTEPS)(x_)
flat = Flatten()(upsamp)
decoded = Dense(units=feature_size, activation = 'relu')(flat)
decoded = Reshape((TIMESTEPS, feature_size))(decoded)

print("shape of decoded ".format(K.int_shape(decoded)))

autoencoder = Model(input_layer, decoded)

ValueError: Negative dimension size caused by subtracting 23 from 1 for 'max_pooling1d/MaxPool' (op: 'MaxPool') with input shapes: [?,1,1,59].

python keras conv-neural-network autoencoder

share|improve this question

asked Mar 6 at 14:24

GuidoGuido

678

add a comment | 

I am a novice in the area of Deep Learning and am willing to build a Conv1D autoencoder for time series with such shapes:

• samples: 200


• timesteps: 23


• features: 178

I am not sure about how should I set the parameters: filters, kernel_size and the layers MaxPooling1D(what's the behaviour?), Flatten(what's the behaviour?), and how should be architecturally designed the simplest Conv1D Autoencoder.

I tried to gather something alone but my version is not working at all:

input_layer = Input(shape=(TIMESTEPS, feature_size))
# ENCODER
x = Conv1D(filters=encoding_dim, kernel_size=TIMESTEPS, activation='relu', padding='valid')(input_layer)
x1 = MaxPooling1D(poolsize=TIMESTEPS)(x)
flat = Flatten()(x1)
encoded = Dense(units=encoding_dim, activation = 'relu')(flat)

print("shape of encoded ".format(K.int_shape(encoded)))

# DECODER 
x_ = Conv1D(encoding_dim, TIMESTEPS, activation='relu', padding='valid')(encoded)
upsamp = UpSampling1D(TIMESTEPS)(x_)
flat = Flatten()(upsamp)
decoded = Dense(units=feature_size, activation = 'relu')(flat)
decoded = Reshape((TIMESTEPS, feature_size))(decoded)

print("shape of decoded ".format(K.int_shape(decoded)))

autoencoder = Model(input_layer, decoded)

ValueError: Negative dimension size caused by subtracting 23 from 1 for 'max_pooling1d/MaxPool' (op: 'MaxPool') with input shapes: [?,1,1,59].

python keras conv-neural-network autoencoder

share|improve this question

asked Mar 6 at 14:24

GuidoGuido

678

input_layer = Input(shape=(TIMESTEPS, feature_size))
# ENCODER
x = Conv1D(filters=encoding_dim, kernel_size=TIMESTEPS, activation='relu', padding='valid')(input_layer)
x1 = MaxPooling1D(poolsize=TIMESTEPS)(x)
flat = Flatten()(x1)
encoded = Dense(units=encoding_dim, activation = 'relu')(flat)

print("shape of encoded ".format(K.int_shape(encoded)))

# DECODER 
x_ = Conv1D(encoding_dim, TIMESTEPS, activation='relu', padding='valid')(encoded)
upsamp = UpSampling1D(TIMESTEPS)(x_)
flat = Flatten()(upsamp)
decoded = Dense(units=feature_size, activation = 'relu')(flat)
decoded = Reshape((TIMESTEPS, feature_size))(decoded)

print("shape of decoded ".format(K.int_shape(decoded)))

autoencoder = Model(input_layer, decoded)

share|improve this question

asked Mar 6 at 14:24

GuidoGuido

678

share|improve this question

asked Mar 6 at 14:24

GuidoGuido

678

asked Mar 6 at 14:24

GuidoGuido

678

asked Mar 6 at 14:24

GuidoGuido

678