A Sparsity Analysis Tool and Library
Project description
hoplite2
a sparsity analysis tool for neural networks
Installation
pip install hoplite2
Usage
There are 2 main classes that are useful in Hoplite2: Spartan and Hoplite.
The Hoplite Class is the main way to use Hoplite2.
from hoplite2 import Hoplite
from tensorflow.keras.applications.vgg16 import VGG16
from tensorflow.keras.preprocessing import image
from tensorflow.keras.applications.vgg16 import preprocess_input
import numpy as np
# keras model to analyze
model = VGG16(
include_top=True,
weights="imagenet",
input_tensor=None,
input_shape=None,
pooling=None,
classes=1000,
classifier_activation="softmax",
)
# preprocess function
def vgg16_preprocess(path):
img = image.load_img(path, target_size=(224, 224))
return preprocess_input(np.expand_dims(image.img_to_array(img), axis=0))
hop = Hoplite(model, vgg16_preprocess, layers=[
"block1_conv2",
"block2_conv2",
"block3_conv3",
"block4_conv3",
"block5_conv3"
])
hop.analyze_file("test.png") # analyzes test.png
hop.output("output.csv") # saves output to file
Spartan implements several useful functions to analyze sparsity of arrays. These functions include:
- compute_average_sparsity(output,equals_zero=lambda x: x == 0)
- consec_1d(arr, hist, equals_zero=lambda x: x == 0)
- consec_row(output, equals_zero=lambda x: x == 0)
- consec_col(output, equals_zero=lambda x: x == 0)
- consec_chan(output, equals_zero=lambda x: x == 0)
- vec_1d(arr, vec_size, hist, equals_zero=lambda x: x == 0):
- vec_3d_row(output, vec_size, equals_zero=lambda x: x == 0):
- vec_3d_col(output, vec_size, equals_zero=lambda x: x == 0):
- vec_3d_chan(output, vec_size, equals_zero=lambda x: x == 0):
equals_zero is a function that takes in a number and returns True if the number is considered zero. This is useful if you wish to look for values that are also close to zero and allows for additional customizablity.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
hoplite2-0.3.0.tar.gz
(9.0 kB
view hashes)
Built Distribution
Close
Hashes for hoplite2-0.3.0-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | bb7827d07f4eb68a8e883213849d9aa8c1439ab067fad067724dca756f97235d |
|
MD5 | aff4fffdb73239ba146c796beae16d2b |
|
BLAKE2b-256 | 311500055f9ff205453a42db42efaa7bdfc62443cc8d850165ebf7b22dc98d23 |