Setup Module

1) htm.core

Pip

n/a

Git

From command line:

• Clone htm.core repo: git clone https://github.com/htm-community/htm.core.git
• CD to htm.core dir: cd htm.core
• Create fresh env: conda create -n htm_env python=3.9.7
• Switch to fresh env: conda activate htm_env
• Install packages: pip install -r requirements.txt
• Run setup.py: python setup.py install

2) htm_streamer

Pip

From command line:

• Install packages: pip install git+https://github.com/gotham29/htm_streamer.git

Git

From command line:

• Clone htm_streamer repo: git clone https://github.com/gotham29/htm_streamer.git
• CD to htm_streamer dir:
  • cd ..
  • cd htm_streamer
• Install packages: pip install -r requirements.txt

3) Run Integration Tests

From command line:

• Run: python tests/integration_tests.py
• Get results: htm_streamer/tests

Quickstart

From Python:

• Import Functions:
  • import os
  • import pandas as pd
  • from htm_streamer.utils.fs import load_config
  • from htm_streamer.pipeline.htm_batch_runner import run_batch
• Load Config & Data:
  • config_path_user = os.path.join(os.getcwd(), 'config', 'config--user_modify.yaml')
  • config_path_model = os.path.join(os.getcwd(), 'config', 'config--model_default.yaml')
  • cfg_user = load_config(config_path_user)
  • cfg_model = load_config(config_path_model)
  • data_path = os.path.join(os.getcwd(), 'data', 'batch', 'sample_timeseries.csv')
  • data = pd.read_csv(data_path)
• Set Config:
  • timestep_tostop_sampling = 40
  • timestep_tostop_learning = 4000
  • timestep_tostop_running = 5000
  • model_for_each_feature = True
  • features_invalid = [ f for f in cfg_user['features'] if f not in data]
  • assert len(features_invalid) == 0, f"features not found --> {sorted(features_invalid)}"
  • cfg_user['timesteps_stop']['sampling'] = timestep_tostop_sampling
  • cfg_user['timesteps_stop']['learning'] = timestep_tostop_learning
  • cfg_user['timesteps_stop']['running'] = timestep_tostop_running
  • cfg_user['models_state']['model_for_each_feature'] = model_for_each_feature
• Train New HTM Models:
  • features_models, features_outputs = run_batch(cfg_user=cfg_user, cfg_model=cfg_model, config_path_user=None, config_path_model=None, learn=True, data=data, iter_print=100, features_models={})
• Run Existing HTM Models:
  • features_models, features_outputs = run_batch(cfg_user=cfg_user, cfg_model=cfg_model, config_path_user=None, config_path_model=None, learn=False, data=data, iter_print=100, features_models=features_models)
• Collect Models and Outputs:
  • f1 = list(cfg_user['features'].keys())[0]
  • f1_model = features_models[ f1 ]
  • f1_anomaly_scores = features_outputs[f1]['anomaly_score']
  • f1_anomaly_liklihoods = features_outputs[f1]['anomaly_likelihood']
  • f1_prediction_counts = features_outputs[f1]['pred_count']