-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathfor_predict_resistivity.yml
32 lines (32 loc) · 2.12 KB
/
for_predict_resistivity.yml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
# - Inferring
# -- Custom neural network
custom_NN: "<module 'my_model' from '../../config/model.py'>" # custom keras model
# -- Directory
# =====
# Automatically read dataset in subdirectories with a specific directory name.
# For example, ""../../data/trial1/testing/resistance/[midpoint]/", "../../data/trial1/testing/resistivity/[section]/
dataset_rootdir: "../../data/trial1"
resistance_dirname: "[midpoint]"
resistivity_dirname: "[section]"
# =====
model_dir: "../../models/trial1" # In this directory, the program will automatically create the predictions directory, read "simulator.pkl" and read "trained_weights" in a subdirectory named weights.
# -- Preprocessing in tf.data.Dataset pipeline. *NOTE: Some operations are data augmentation.*
preprocess: # Add_noise is implemented earlier than log_transform
add_noise: # Because we add random noise every time, this operation is data augmentation.
perform: False # {True, False}. Whether to perform add_noise.
kwargs:
scale: 0.1 # Noise added to element is proportional to this value.
noise_type: "normal" # {'normal', 'uniform'}
log_transform:
perform: False # {True, False}. Whether to perform log_transform.
kwargs:
inverse: False # {True, False}. Whether to perform an inverse transformation.
inplace: True # {True, False}. Whether to use inplace mode.
to_midpoint: # Reshape "inputs" tensor to midpoint image. shape = (accumulated number of same midpoint, number of midpoint, 1)
perform: False # {True, False}. Whether to perform to_midpoint. *NOTE: Don't use `to_midpoint` and `to_txrx` at the same time*
to_txrx: # Reshape "inputs" tensor to Tx-Rx image. shape = (number of Tx pair, number of Rx pair, 1)
perform: False # {True, False}. Whether to perform to_TxRx. *NOTE: Don't use `to_midpoint` and `to_txrx` at the same time*
to_section: # Reshape "target" tensor to section image. shape = (number of cell center mesh in z direction, number of cell center mesh in x direction, 1)
perform: False # {True, False}. Whether to perform to_section.
# -- Hyper parameters
num_gpu: 2 # Number of gpu