Please enable JavaScript.
Coggle requires JavaScript to display documents.
KERAS (LAYERS ("nodes" (Each call, a NODE is created in this…
KERAS
LAYERS
In Multi-Input, Output, we have to define LOSS_WEIGHTS
MultiInput, MultiOutput: When these models share multilayers. Each pair (Input, Output) will define a DAG for training
"nodes"
Each call, a NODE is created in this layer to LINK the input and the output
a layers would be called many times
each NODE is INDEXED 0, 1, 2 ...
layers.get_output_at(index)
Model M is also callable; Just call M(input)
Callable; Input is a Tensor, Output is a Tensor also
kkk
MODELS
SEQUENTIAL
the SIMPLEST type of models.
Steps to build a model
add layers from Keras
compile the instance by setting optimizer, loss function, etc
declare an instance of Sequential(......)
declare Input and Labels of type Numpy arrays
fit and predict
a linear stack of layers
MODEL API
Gen Info
common functions
model.summary()
model.get_config() returns a dict of configuration params
model.from_config(config) to instantiate a mode from a configuration dict
model.get_weights() returns LIST of tensor weights
model.set_weights(weights) with weights as List of tensor weights
model.to_json(), model.to_yaml(): save model ARCHITECTURE into json and yaml string, respectively.
model_from_json(json_string), model_from_yaml(yaml_string) to construct a model ARCHITECTURE from json and yaml string
2 types
Model Class with functional API
Sequential Model
Losses, Optimizers, Metrics, Backend, Initializers, Regularizers, Constraints, Visualization, etc.