Loading a Trained Neuron Array
In this tutorial we will demonstrate how to load a neuron array directly into the pipeline in order to create pipelines based off of pre-trained models
Try It Yourself
You can download the Notebook file here to follow along with this tutorial in your own environment.
[6]:
import pandas as pd
from sensiml import SensiML
client = SensiML()
client.project ='Activity Case Study'
client.pipeline = 'Activity_Pipeline'
[7]:
sensor_columns = client.project.columns()
df = pd.read_csv('data/activity_data_medium.csv')
df['Subject'] = df['Subject'].apply(lambda x: int(x[1:]))
client.upload_dataframe('activity_data.csv', df)
df.head()
Uploading file "activity_data.csv" to SensiML Cloud.
[7]:
| Subject | Activity | AccelerometerX | AccelerometerY | AccelerometerZ | |
|---|---|---|---|---|---|
| 0 | 1 | 0 | -317 | -3000 | 925 |
| 1 | 1 | 0 | -284 | -2968 | 903 |
| 2 | 1 | 0 | -243 | -2987 | 933 |
| 3 | 1 | 0 | -193 | -3051 | 936 |
| 4 | 1 | 0 | -150 | -3059 | 915 |
Neuron Array Format
The following is the format of a neuron array for a trained model.
[8]:
neuron_array = [{u'AIF': 1,
u'Category': 2,
u'Context': 1,
u'Identifier': 1,
u'Vector': [0, 0, 0, 0, 0, 124, 0, 0]},
{u'AIF': 1,
u'Category': 2,
u'Context': 1,
u'Identifier': 2,
u'Vector': [166, 31, 172, 138, 93, 31, 254, 35]},
{u'AIF': 143,
u'Category': 3,
u'Context': 1,
u'Identifier': 3,
u'Vector': [246, 232, 243, 68, 241, 156, 72, 16]},
{u'AIF': 1,
u'Category': 3,
u'Context': 1,
u'Identifier': 4,
u'Vector': [134, 171, 163, 42, 167, 114, 19, 254]},
{u'AIF': 349,
u'Category': 4,
u'Context': 1,
u'Identifier': 5,
u'Vector': [118, 165, 147, 184, 156, 191, 2, 0]},
{u'AIF': 449,
u'Category': 1,
u'Context': 1,
u'Identifier': 6,
u'Vector': [97, 78, 107, 43, 92, 106, 56, 4]},
{u'AIF': 354,
u'Category': 4,
u'Context': 1,
u'Identifier': 7,
u'Vector': [131, 184, 165, 53, 174, 128, 2, 3]}]
Create a Pipeline
Create a simple pipeline.
[9]:
client.pipeline.reset()
client.pipeline.set_columns(data_columns=['AccelerometerY'], group_columns=['Activity','Subject'], label_column='Activity')
client.pipeline.set_input_data('activity_data.csv', data_columns=['AccelerometerY'])
client.pipeline.add_transform('Windowing')
client.pipeline.add_feature_generator(["Mean", 'Standard Deviation', 'Skewness', 'Kurtosis',
'25th Percentile', '75th Percentile', '100th Percentile',
'Zero Crossing Rate'],
function_defaults = {"columns":[u'AccelerometerY']})
client.pipeline.add_transform('Min Max Scale')
Load the Neuron Array
Here we will use the training algorithm “Load Neuron Array” which will load the neurons directly into the hardware simulator and run the cross-fold validation on the model. Note: No training occurs here, we are simply using the provided neurons.
[10]:
client.pipeline.set_training_algorithm('Load Neuron Array', params = {'neuron_array':neuron_array})
client.pipeline.set_validation_method('Stratified K-Fold Cross-Validation', params={'number_of_folds':5})
client.pipeline.set_classifier('PME', params={"classification_mode":'RBF','distance_mode':'L1'})
client.pipeline.set_tvo()
[11]:
results, stats = client.pipeline.execute()
Executing Pipeline with Steps:
------------------------------------------------------------------------
0. Name: activity_data.csv Type: featurefile
------------------------------------------------------------------------
------------------------------------------------------------------------
1. Name: Windowing Type: segmenter
------------------------------------------------------------------------
------------------------------------------------------------------------
2. Name: generator_set Type: generatorset
------------------------------------------------------------------------
------------------------------------------------------------------------
3. Name: Min Max Scale Type: transform
------------------------------------------------------------------------
------------------------------------------------------------------------
4. Name: tvo Type: tvo
------------------------------------------------------------------------
Classifier: PME
classification_mode: RBF
distance_mode: L1
max_aif: 400
min_aif: 25
num_channels: 1
reinforcement_learning: False
reserved_patterns: 0
Training Algo: Load Neuron Array
class_map: {}
neuron_array: [{'AIF': 1, 'Category': 2, 'Context': 1, 'Identifier': 1, 'Vector': [0, 0, 0, 0, 0, 124, 0, 0]}, {'AIF': 1, 'Category': 2, 'Context': 1, 'Identifier': 2, 'Vector': [166, 31, 172, 138, 93, 31, 254, 35]}, {'AIF': 143, 'Category': 3, 'Context': 1, 'Identifier': 3, 'Vector': [246, 232, 243, 68, 241, 156, 72, 16]}, {'AIF': 1, 'Category': 3, 'Context': 1, 'Identifier': 4, 'Vector': [134, 171, 163, 42, 167, 114, 19, 254]}, {'AIF': 349, 'Category': 4, 'Context': 1, 'Identifier': 5, 'Vector': [118, 165, 147, 184, 156, 191, 2, 0]}, {'AIF': 449, 'Category': 1, 'Context': 1, 'Identifier': 6, 'Vector': [97, 78, 107, 43, 92, 106, 56, 4]}, {'AIF': 354, 'Category': 4, 'Context': 1, 'Identifier': 7, 'Vector': [131, 184, 165, 53, 174, 128, 2, 3]}]
Validation Method: Stratified K-Fold Cross-Validation
number_of_folds: 5
shuffle: True
test_size: 0.0
------------------------------------------------------------------------
Results Retrieved... Execution Time: 0 min. 15 sec.
[12]:
results.summarize()
TRAINING ALGORITHM: Load Neuron Array
VALIDATION METHOD: Stratified K-Fold Cross-Validation
CLASSIFIER: PME
AVERAGE METRICS:
F1_SCORE: 13.8 std: 1.60
PRECISION: 28.5 std: 11.54
SENSITIVITY: 25.6 std: 0.67
--------------------------------------
STRATIFIED K-FOLD CROSS-VALIDATION MODEL RESULTS : SET VALIDATION
MODEL INDEX: Fold 1
F1_SCORE: train: 13.20 validation: 16.37
SENSITIVITY: train: 25.33 validation: 26.75
MODEL INDEX: Fold 3
F1_SCORE: train: 13.90 validation: 13.65
SENSITIVITY: train: 25.66 validation: 25.44
MODEL INDEX: Fold 4
F1_SCORE: train: 13.87 validation: 13.80
SENSITIVITY: train: 25.66 validation: 25.44
MODEL INDEX: Fold 0
F1_SCORE: train: 13.96 validation: 13.42
SENSITIVITY: train: 25.66 validation: 25.43
MODEL INDEX: Fold 2
F1_SCORE: train: 14.33 validation: 11.96
SENSITIVITY: train: 25.76 validation: 25.00