Skip to content

Latest commit

 

History

History
57 lines (42 loc) · 3.43 KB

README.md

File metadata and controls

57 lines (42 loc) · 3.43 KB
Raw

2. Train Model

In this module, we train ML models to predict phenotypic class from cell features.

In train_multi_class_models.ipynb, we train models to predict the phenotypic class of the cell features from 15 possible classes (anaphase, metaphase, apoptosis, etc). In train_single_class_models.ipynb, we train models to predict whether the cell features are from a particular phenotypic class or not. Thus, we create a set of models to predict a "yes" or "no" for each of the 15 phenotypic classes used in the multi-class models.

We use sklearn.linear_model.LogisticRegression for our machine learning models. We use the following parameters for our each Logisic Regression model:

  • penalty='elasticnet': Use elasticnet as the penalty for our model. Elastic-Net regularization is a combination of L1 and L2 regularization methods. The mixing of these two methods is determined by the l1_ratio parameter which we optimize later.
  • solver='saga': We use the saga solver as this is the only solver that supports Elastic-Net regularization.
  • max_iter=100: Set the maximum number of iterations for solver to converge. 100 iterations allows the solver to maximize performance without completing unnecessary iterations.

We use sklearn.model_selection.GridSearchCV to perform an exhaustive search for the parameters below. This searches for parameters that maximize the weighted F1 score of the model. We optimize weighted F1 score because this metric measures model precision and recall and accounts for label imbalance (see sklearn.metrics.f1_score for more details).

  • l1_ratio: Elastic-Net mixing parameter. Used to combine L1 and L2 regularization methods. We search over the following parameters: [0. , 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1. ]
  • C: Inversely proportional to regularization strength. We search over the following parameters: [1.e-03, 1.e-02, 1.e-01, 1.e+00, 1.e+01, 1.e+02, 1.e+03]

We train models for each combination of the following model types, feature, balance, and dataset types:

  • model_types: final, shuffled_baseline
    • Which version of features the model is trained on. For shuffled_baseline, each column of the feature data is shuffled independently to create a shuffled baseline for comparison.
  • feature_types: CP, DP, CP_and_DP, CP_zernike_only, CP_areashape_only
    • Which features to use for training.
  • balance_types: balanced, unbalanced
    • Whether or not to balance class_weight of each model when training.
  • dataset_types: ic, no_ic
    • Which mitocheck_data dataset to use for feature training. We have datasets extracted with and without illumination correction.

The notebooks save each model in models/.

Step 1: Train Model

Use the commands below to train the ML model:

# Make sure you are located in 2.train_model
cd 2.train_model

# Activate phenotypic_profiling conda environment
conda activate phenotypic_profiling

# Train model
bash train_model.sh

Results

The weighted F1 score of the best estimators for the grid searches are located in train_model.ipynb.