RSGISLib LightGBM Pixel Classification

LightGBM (https://lightgbm.readthedocs.io) is an alternative library to scikit-learn which has specialist implementation of Gradient Boosted Decision Tree (GBDT), but it also implements random forests, Dropouts meet Multiple Additive Regression Trees (DART), and Gradient Based One-Side Sampling (Goss).

When considering ensemble learning, there are two primary methods: bagging and boosting. Bagging involves the training of many independent models and combines their predictions through some form of aggregation (averaging, voting etc.). An example of a bagging ensemble is a Random Forest.

Boosting instead trains models sequentially, where each model learns from the errors of the previous model. Starting with a weak base model, models are trained iteratively, each adding to the prediction of the previous model to produce a strong overall prediction. In the case of gradient boosted decision trees, successive models are found by applying gradient descent in the direction of the average gradient, calculated with respect to the error residuals of the loss function, of the leaf nodes of previous models.

See also

For an an academic paper see: Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., Liu, T.LightGBM: A Highly Efficient Gradient Boosting Decision Tree. 31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.

LightGBM is a binary classifier (i.e., separates two classes, e.g., mangroves and other) but it has a multi-class mode which applies a number of binary classification to produce a multi-class classification result.

Steps to applying a LightGBM Classification:

Extract training

Split training: Training, Validation, Testing

Train Classifier and Optimise Hyperparameters

Apply Classifier

However, fist we’ll create a couple of directories for our outputs and intermediary files:

import os

out_dir = "baseline_cls_lgbm"
if not os.path.exists(out_dir):
    os.mkdir(out_dir)

tmp_dir = "tmp_lgbm"
if not os.path.exists(tmp_dir):
    os.mkdir(tmp_dir)

We will also define the input file path and the list ImageBandInfo objects, which specifies which images and bands are used for the analysis:

import rsgislib.imageutils

input_img = "./LS5TM_19970716_vmsk_mclds_topshad_rad_srefdem_stdsref_subset.tif"

imgs_info = []
imgs_info.append(
    rsgislib.imageutils.ImageBandInfo(
        file_name=input_img, name="ls97", bands=[1, 2, 3, 4, 5, 6]
    )
)

When applying a classifier a mask image needs to be provided where a pixel value within that mask specifying which pixels should be classified. While defining the input image we can also define that valid mask image using the rsgislib.imageutils.gen_valid_mask function, which simply creates a mask of pixels which are not ‘no data’:

vld_msk_img = os.path.join(out_dir, "LS5TM_19970716_vmsk.kea")
rsgislib.imageutils.gen_valid_mask(
    input_img, output_img=vld_msk_img, gdalformat="KEA", no_data_val=0.0
)

To define training a raster with a unique value for each class, or multiple binary rasters one for each class. Commonly the training regions might be defined using a vector layer which would require rasterising:

import rsgislib.vectorutils.createrasters

mangrove_vec_file = "./training/mangroves.geojson"
mangrove_vec_lyr = "mangroves"
mangrove_smpls_img = os.path.join(tmp_dir, "mangrove_smpls.kea")
rsgislib.vectorutils.createrasters.rasterise_vec_lyr(
    vec_file=mangrove_vec_file,
    vec_lyr=mangrove_vec_lyr,
    input_img=input_img,
    output_img=mangrove_smpls_img,
    gdalformat="KEA",
    burn_val=1,
)

other_terrestrial_vec_file = "./training/other_terrestrial.geojson"
other_terrestrial_vec_lyr = "other_terrestrial"
other_terrestrial_smpls_img = os.path.join(tmp_dir, "other_terrestrial_smpls.kea")
rsgislib.vectorutils.createrasters.rasterise_vec_lyr(
    vec_file=other_terrestrial_vec_file,
    vec_lyr=other_terrestrial_vec_lyr,
    input_img=input_img,
    output_img=other_terrestrial_smpls_img,
    gdalformat="KEA",
    burn_val=1,
)

water_vec_file = "./training/water.geojson"
water_vec_lyr = "water"
water_smpls_img = os.path.join(tmp_dir, "water_smpls.kea")
rsgislib.vectorutils.createrasters.rasterise_vec_lyr(
    vec_file=water_vec_file,
    vec_lyr=water_vec_lyr,
    input_img=input_img,
    output_img=water_smpls_img,
    gdalformat="KEA",
    burn_val=1,
)

To extract the image pixel values, which are stored within a HDF5 file (see https://portal.hdfgroup.org/display/HDF5/HDF5 for more information) the following functions are used. To define the images and associated bands to be used for the classification and therefore values need to be extracted then a list of rsgislib.imageutils.ImageBandInfo classes needs to be provided:

import rsgislib.zonalstats

mangrove_all_smpls_h5_file = os.path.join(out_dir, "mangrove_all_smpls.h5")
rsgislib.zonalstats.extract_zone_img_band_values_to_hdf(
    imgs_info,
    in_msk_img=mangrove_smpls_img,
    out_h5_file=mangrove_all_smpls_h5_file,
    mask_val=1,
    datatype=rsgislib.TYPE_16UINT,
)

other_terrestrial_all_smpls_h5_file = os.path.join(
    out_dir, "other_terrestrial_all_smpls.h5"
)
rsgislib.zonalstats.extract_zone_img_band_values_to_hdf(
    imgs_info,
    in_msk_img=other_terrestrial_smpls_img,
    out_h5_file=other_terrestrial_all_smpls_h5_file,
    mask_val=1,
    datatype=rsgislib.TYPE_16UINT,
)

water_all_smpls_h5_file = os.path.join(out_dir, "water_all_smpls.h5")
rsgislib.zonalstats.extract_zone_img_band_values_to_hdf(
    imgs_info,
    in_msk_img=water_smpls_img,
    out_h5_file=water_all_smpls_h5_file,
    mask_val=1,
    datatype=rsgislib.TYPE_16UINT,
)

If training data is extracted from multiple input images then it will need to be merged using the following function. In this case we’ll merge the water and terrestrial samples and use the merged class to create a mangrove binary classifier:

other_all_smpls_h5_file = os.path.join(out_dir, "other_all_smpls.h5")
rsgislib.zonalstats.merge_extracted_hdf5_data(
    h5_files=[other_terrestrial_all_smpls_h5_file, water_all_smpls_h5_file],
    out_h5_file=other_all_smpls_h5_file,
    datatype=rsgislib.TYPE_16UINT,
)

To split the extracted samples into a training, validation and testing sets you can use the rsgislib.classification.split_sample_train_valid_test function. Note, this function is also used to standardise the number of samples used to train the classifier so the training data are balanced:

import rsgislib.classification

mangrove_train_smpls_h5_file = os.path.join(out_dir, "mangrove_train_smpls.h5")
mangrove_valid_smpls_h5_file = os.path.join(out_dir, "mangrove_valid_smpls.h5")
mangrove_test_smpls_h5_file = os.path.join(out_dir, "mangrove_test_smpls.h5")
rsgislib.classification.split_sample_train_valid_test(
    in_h5_file=mangrove_all_smpls_h5_file,
    train_h5_file=mangrove_train_smpls_h5_file,
    valid_h5_file=mangrove_valid_smpls_h5_file,
    test_h5_file=mangrove_test_smpls_h5_file,
    test_sample=10000,
    valid_sample=10000,
    train_sample=35000,
    rnd_seed=42,
    datatype=rsgislib.TYPE_16UINT,
)


other_terrestrial_train_smpls_h5_file = os.path.join(
    out_dir, "other_terrestrial_train_smpls.h5"
)
other_terrestrial_valid_smpls_h5_file = os.path.join(
    out_dir, "other_terrestrial_valid_smpls.h5"
)
other_terrestrial_test_smpls_h5_file = os.path.join(
    out_dir, "other_terrestrial_test_smpls.h5"
)
rsgislib.classification.split_sample_train_valid_test(
    in_h5_file=other_terrestrial_all_smpls_h5_file,
    train_h5_file=other_terrestrial_train_smpls_h5_file,
    valid_h5_file=other_terrestrial_valid_smpls_h5_file,
    test_h5_file=other_terrestrial_test_smpls_h5_file,
    test_sample=10000,
    valid_sample=10000,
    train_sample=35000,
    rnd_seed=42,
    datatype=rsgislib.TYPE_16UINT,
)


water_train_smpls_h5_file = os.path.join(out_dir, "water_train_smpls.h5")
water_valid_smpls_h5_file = os.path.join(out_dir, "water_valid_smpls.h5")
water_test_smpls_h5_file = os.path.join(out_dir, "water_test_smpls.h5")
rsgislib.classification.split_sample_train_valid_test(
    in_h5_file=water_all_smpls_h5_file,
    train_h5_file=water_train_smpls_h5_file,
    valid_h5_file=water_valid_smpls_h5_file,
    test_h5_file=water_test_smpls_h5_file,
    test_sample=10000,
    valid_sample=10000,
    train_sample=35000,
    rnd_seed=42,
    datatype=rsgislib.TYPE_16UINT,
)


other_train_smpls_h5_file = os.path.join(out_dir, "other_train_smpls.h5")
other_valid_smpls_h5_file = os.path.join(out_dir, "other_valid_smpls.h5")
other_test_smpls_h5_file = os.path.join(out_dir, "other_test_smpls.h5")
rsgislib.classification.split_sample_train_valid_test(
    in_h5_file=other_all_smpls_h5_file,
    train_h5_file=other_train_smpls_h5_file,
    valid_h5_file=other_valid_smpls_h5_file,
    test_h5_file=other_test_smpls_h5_file,
    test_sample=10000,
    valid_sample=10000,
    train_sample=35000,
    rnd_seed=42,
    datatype=rsgislib.TYPE_16UINT,
)

Note

Training samples are used to train the classifier. Validation samples are used to test the accuracy of the classifier during the parameter optimisation process and are therefore part of the training process and not independent. Testing samples completely independent of the training process and are used as an independent sample to test the overall accuracy of the classifier.

Apply a LightGBM Binary Classifier

To train a single binary classifier you need to use the following function:

import rsgislib.classification.classlightgbm

cls_bin_mdl_file = os.path.join(out_dir, "lgbm_mng_bin_mdl.txt")
rsgislib.classification.classlightgbm.train_opt_lightgbm_binary_classifier(
    out_mdl_file=cls_bin_mdl_file,
    cls1_train_file=mangrove_train_smpls_h5_file,
    cls1_valid_file=mangrove_valid_smpls_h5_file,
    cls1_test_file=mangrove_test_smpls_h5_file,
    cls2_train_file=other_train_smpls_h5_file,
    cls2_valid_file=other_valid_smpls_h5_file,
    cls2_test_file=other_test_smpls_h5_file,
    unbalanced=False,
    op_mthd=rsgislib.OPT_MTHD_BAYESOPT,
    n_opt_iters=100,
    rnd_seed=42,
    n_threads=1,
    scale_pos_weight=None,
    early_stopping_rounds=None,
    num_iterations=100,
    max_n_leaves=50,
    learning_rate=0.1,
    mdl_cls_obj=None,
    out_params_file=None,
)

To apply the binary classifier use the following function:

cls_score_img = os.path.join(out_dir, "LS5TM_19970716_bin_cls_score_img.kea")
out_class_img = os.path.join(out_dir, "LS5TM_19970716_bin_cls_img.kea")
rsgislib.classification.classlightgbm.apply_lightgbm_binary_classifier(
    model_file=cls_bin_mdl_file,
    in_msk_img=vld_msk_img,
    img_msk_val=1,
    img_file_info=imgs_info,
    out_score_img=cls_score_img,
    gdalformat="KEA",
    out_class_img=out_class_img,
    class_thres=5000,
)

Note

Class probability values are multipled by 10,000 so a threshold of 5000 is really 0.5.

Apply a LightGBM Multi-Class Classifier

To train a multi-class classifier you first need to specify the reference samples as a dict of rsgislib.classification.ClassInfoObj objects:

import rsgislib.classification
import rsgislib.classification.classlightgbm

cls_info_dict = dict()
cls_info_dict["Mangrove"] = rsgislib.classification.ClassInfoObj(
    id=0,
    out_id=1,
    train_file_h5=mangrove_train_smpls_h5_file,
    test_file_h5=mangrove_test_smpls_h5_file,
    valid_file_h5=mangrove_valid_smpls_h5_file,
    red=0,
    green=255,
    blue=0,
)
cls_info_dict["Other Terrestrial"] = rsgislib.classification.ClassInfoObj(
    id=1,
    out_id=2,
    train_file_h5=other_terrestrial_train_smpls_h5_file,
    test_file_h5=other_terrestrial_test_smpls_h5_file,
    valid_file_h5=other_terrestrial_valid_smpls_h5_file,
    red=100,
    green=100,
    blue=100,
)
cls_info_dict["Water"] = rsgislib.classification.ClassInfoObj(
    id=2,
    out_id=3,
    train_file_h5=water_train_smpls_h5_file,
    test_file_h5=water_test_smpls_h5_file,
    valid_file_h5=water_valid_smpls_h5_file,
    red=0,
    green=0,
    blue=255,
)

You can then train a multi-class lightgbm classifier using the following function:

import rsgislib.classification.classlightgbm

cls_mcls_mdl_file = os.path.join(out_dir, "lgbm_mng_mcls_mdl.txt")
rsgislib.classification.classlightgbm.train_opt_lightgbm_multiclass_classifier(
    out_mdl_file=cls_mcls_mdl_file,
    cls_info_dict=cls_info_dict,
    unbalanced=False,
    op_mthd=rsgislib.OPT_MTHD_BAYESOPT,
    n_opt_iters=100,
    rnd_seed=42,
    n_threads=1,
    early_stopping_rounds=None,
    num_iterations=100,
    max_n_leaves=50,
    learning_rate=0.1,
    mdl_cls_obj=None,
    out_params_file=None,
    out_info_file=None,
)

To apply the multi-class classifier use the following function:

out_class_img = os.path.join(out_dir, "LS5TM_19970716_mcls_img.kea")
rsgislib.classification.classlightgbm.apply_lightgbm_multiclass_classifier(
    model_file=cls_mcls_mdl_file,
    cls_info_dict=cls_info_dict,
    in_msk_img=vld_msk_img,
    img_msk_val=1,
    img_file_info=imgs_info,
    out_class_img=out_class_img,
    gdalformat="KEA",
    class_clr_names=True,
)

Note

Within the rsgislib.classification.ClassInfoObj class you need to provide an id and out_id value. The id must start from zero and be consecutive while the out_id will be used as the pixel value for the output classification image and can be any integer value.

Binary Classification Functions

rsgislib.classification.classlightgbm.optimise_lightgbm_binary_classifier(out_params_file: str, cls1_train_file: str, cls1_valid_file: str, cls2_train_file: str, cls2_valid_file: str, unbalanced: bool = False, op_mthd: int = 1, n_opt_iters: int = 100, rnd_seed: int = None, n_threads: int = 1, scale_pos_weight: float = None, early_stopping_rounds: int = None, num_iterations: int = 100, max_n_leaves: int = 50, learning_rate: float = 0.1, mdl_cls_obj=None)

A function which performs a hyper-parameter optimisation for a binary lightgbm classifier. Class 1 is the class which you are interested in and Class 2 is the ‘other class’.

You have the option of using the bayes_opt (Default), optuna or skopt optimisation libraries. Before 5.1.0 skopt was the only option but this no longer appears to be maintained so the other options have been added.

Parameters:

out_params_file – The output JSON file with the identified parameters
cls1_train_file – File path to the HDF5 file with the training samples for class 1
cls1_valid_file – File path to the HDF5 file with the validation samples for class 1
cls2_train_file – File Path to the HDF5 file with the training samples for class 2
cls2_valid_file – File path to the HDF5 file with the validation samples for class 2
unbalanced – Boolean (Default: False) specifying whether the training data is unbalanced (i.e., a different number of samples for each class).
op_mthd – The method used to optimise the parameters. Default: rsgislib.OPT_MTHD_BAYESOPT
n_opt_iters – The number of iterations (Default 100) used for the optimisation. This parameter is ignored for skopt. For bayes_opt there is a minimum of 10 and these are added to that minimum so Default is therefore 110. For optuna this is the number of iterations used.
rnd_seed – A random seed for the optimisation. Default None. If None there a different seed will be used each time the function is run.
n_threads – The number of threads used by lightgbm
scale_pos_weight – Optional, default is None. If None then a value will automatically be calculated. Parameter used to balance imbalanced training data.
early_stopping_rounds – If not None then activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training.
num_iterations – The number of boosting iterations (Default: 100)
max_n_leaves – The upper limited used within the optimisation search for the maximum number of leaves used within the model. Default: 50.
learning_rate – Default 0.1 (constraint > 0.0) controlling the shrinkage rate
mdl_cls_obj – An optional (Default None) lightgbm model which will be used as the basis model from which training will be continued (i.e., transfer learning).

rsgislib.classification.classlightgbm.train_lightgbm_binary_classifier(out_mdl_file: str, cls_params_file: str, cls1_train_file: str, cls1_valid_file: str, cls1_test_file: str, cls2_train_file: str, cls2_valid_file: str, cls2_test_file: str, unbalanced: bool = False, n_threads: int = 1, scale_pos_weight: float = None, early_stopping_rounds: int = None, num_iterations: int = 100, learning_rate: float = 0.1, mdl_cls_obj=None)

A function which trains a binary lightgbm model using the parameters provided within a JSON file. The JSON file must provide values for the following parameters:

max_depth

num_leaves

min_data_in_leaf

lambda_l1

lambda_l2

feature_fraction

bagging_fraction

min_split_gain

min_child_weight

reg_alpha

reg_lambda

Parameters:

out_mdl_file – The file path for the output lightgbm (*.txt) model which can be loaded to perform a classification.
cls_params_file – The file path to the JSON file with the classifier parameters.
cls1_train_file – File path to the HDF5 file with the training samples for class 1
cls1_valid_file – File path to the HDF5 file with the validation samples for class 1
cls1_test_file – File path to the HDF5 file with the testing samples for class 1
cls2_train_file – File path to the HDF5 file with the training samples for class 2
cls2_valid_file – File path to the HDF5 file with the validation samples for class 2
cls2_test_file – File path to the HDF5 file with the testing samples for class 2
unbalanced – Boolean (Default: False) specifying whether the training data is unbalanced (i.e., a different number of samples for each class).
n_threads – The number of threads used by lightgbm
scale_pos_weight – Optional, default is None. If None then a value will automatically be calculated. Parameter used to balance imbalanced training data.
early_stopping_rounds – If not None then activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training.
num_iterations – The number of boosting iterations (Default: 100)
learning_rate – Default 0.1 (constraint > 0.0) controlling the shrinkage rate
mdl_cls_obj – An optional (Default None) lightgbm model which will be used as the basis model from which training will be continued (i.e., transfer learning).

rsgislib.classification.classlightgbm.train_opt_lightgbm_binary_classifier(out_mdl_file: str, cls1_train_file: str, cls1_valid_file: str, cls1_test_file: str, cls2_train_file: str, cls2_valid_file: str, cls2_test_file: str, unbalanced: bool = False, op_mthd: int = 1, n_opt_iters: int = 100, rnd_seed: int = None, n_threads: int = 1, scale_pos_weight: float = None, early_stopping_rounds: int = None, num_iterations: int = 100, max_n_leaves: int = 50, learning_rate: float = 0.1, mdl_cls_obj=None, out_params_file: str = None)

A function which performs a hyper-parameter optimisation for a binary lightgbm classifier and then trains a model saving the model for future use. Class 1 is the class which you are interested in and Class 2 is the ‘other class’.

You have the option of using the bayes_opt (Default), optuna or skopt optimisation libraries. Before 5.1.0 skopt was the only option but this no longer appears to be maintained so the other options have been added.

Parameters:

out_mdl_file – The file path for the output lightgbm (*.txt) model which can be loaded to perform a classification.
cls1_train_file – File path to the HDF5 file with the training samples for class 1
cls1_valid_file – File path to the HDF5 file with the validation samples for class 1
cls1_test_file – File path to the HDF5 file with the testing samples for class 1
cls2_train_file – File path to the HDF5 file with the training samples for class 2
cls2_valid_file – File path to the HDF5 file with the validation samples for class 2
cls2_test_file – File path to the HDF5 file with the testing samples for class 2
unbalanced – Boolean (Default: False) specifying whether the training data is unbalanced (i.e., a different number of samples for each class).
op_mthd – The method used to optimise the parameters. Default: rsgislib.OPT_MTHD_BAYESOPT
n_opt_iters – The number of iterations (Default 100) used for the optimisation. This parameter is ignored for skopt. For bayes_opt there is a minimum of 10 and these are added to that minimum so Default is therefore 110. For optuna this is the number of iterations used.
rnd_seed – A random seed for the optimisation. Default None. If None there a different seed will be used each time the function is run.
n_threads – The number of threads used by lightgbm
scale_pos_weight – Optional, default is None. If None then a value will automatically be calculated. Parameter used to balance imbalanced training data.
early_stopping_rounds – If not None then activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training.
num_iterations – The number of boosting iterations (Default: 100)
max_n_leaves – The upper limited used within the optimisation search for the maximum number of leaves used within the model. Default: 50.
learning_rate – Default 0.1 (constraint > 0.0) controlling the shrinkage rate
mdl_cls_obj – An optional (Default None) lightgbm model which will be used as the basis model from which training will be continued (i.e., transfer learning).
out_params_file – The output JSON file with the identified parameters. If None (default) then no file is outputted.

rsgislib.classification.classlightgbm.apply_lightgbm_binary_classifier(model_file: str, in_msk_img: str, img_msk_val: int, img_file_info: List[ImageBandInfo], out_score_img: str, gdalformat: str = 'KEA', out_class_img: str = None, class_thres: int = 5000)

A function for applying a trained binary lightgbm model to a image or stack of image files.

Parameters:

model_file – a trained lightgbm binary model which can be loaded with lgb.Booster(model_file=model_file).
in_msk_img – is an image file providing a mask to specify where should be classified. Simplest mask is all the valid data regions (rsgislib.imageutils.gen_valid_mask)
img_msk_val – the pixel value within the in_msk_img to limit the region to which the classification is applied. Can be used to create a hierarchical classification.
img_file_info – a list of rsgislib.imageutils.ImageBandInfo objects to identify which images and bands are to be used for the classification so it adheres to the training data.
out_score_img – output image file with the classification softmax score. Note. this image is scaled by multiplying by 10000 therefore the range is between 0-10000.
gdalformat – The output image format (Default: KEA).
out_class_img – Optional output image which will contain the hard classification, defined with a threshold on the softmax score image.
class_thres – The threshold used to define the hard classification. Default is 5000 (i.e., softmax score of 0.5).

Multi-Class Classification Functions

rsgislib.classification.classlightgbm.optimise_lightgbm_multiclass_classifier(out_params_file: str, cls_info_dict: Dict[str, ClassInfoObj], unbalanced: bool = False, op_mthd: int = 1, n_opt_iters: int = 100, rnd_seed: int = None, n_threads: int = 1, early_stopping_rounds: int = None, num_iterations: int = 100, max_n_leaves: int = 50, learning_rate: float = 0.1, mdl_cls_obj=None)

A function which performs a hyper-parameter optimisation for a multi-class lightgbm classifier.

You have the option of using the bayes_opt (Default), optuna or skopt optimisation libraries. Before 5.1.0 skopt was the only option but this no longer appears to be maintained so the other options have been added.

Parameters:

out_params_file – The output JSON file with the identified parameters
cls_info_dict – a dict where the key is string with class name of ClassInfoObj objects defining the training data.
unbalanced – Boolean (Default: False) specifying whether the training data is unbalanced (i.e., a different number of samples for each class).
op_mthd – The method used to optimise the parameters. Default: rsgislib.OPT_MTHD_BAYESOPT
n_opt_iters – The number of iterations (Default 100) used for the optimisation. This parameter is ignored for skopt. For bayes_opt there is a minimum of 10 and these are added to that minimum so Default is therefore 110. For optuna this is the number of iterations used.
rnd_seed – A random seed for the optimisation. Default None. If None there a different seed will be used each time the function is run.
n_threads – The number of threads used by lightgbm
early_stopping_rounds – If not None then activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training.
num_iterations – The number of boosting iterations (Default: 100)
max_n_leaves – The upper limited used within the optimisation search for the maximum number of leaves used within the model. Default: 50.
learning_rate – Default 0.1 (constraint > 0.0) controlling the shrinkage rate
mdl_cls_obj – An optional (Default None) lightgbm model which will be used as the basis model from which training will be continued (i.e., transfer learning).

rsgislib.classification.classlightgbm.train_lightgbm_multiclass_classifier(params_file: str, out_mdl_file: str, cls_info_dict: Dict[str, ClassInfoObj], unbalanced: bool = False, n_threads: int = 1, early_stopping_rounds: int = None, num_iterations: int = 100, learning_rate: float = 0.1, mdl_cls_obj=None, out_info_file: str = None)

A function which trains a multiclass lightgbm model using the parameters provided within a JSON file. The JSON file must provide values for the following parameters:

max_depth

num_leaves

min_data_in_leaf

lambda_l1

lambda_l2

feature_fraction

bagging_fraction

min_split_gain

min_child_weight

reg_alpha

reg_lambda

Parameters:

params_file – The file path to the JSON file with the classifier parameters.
out_mdl_file – The file path for the output lightgbm (*.txt) model which can be loaded to perform a classification.
cls_info_dict – a dict where the key is string with class name of ClassInfoObj objects defining the training data.
unbalanced – Boolean (Default: False) specifying whether the training data is unbalanced (i.e., a different number of samples for each class).
n_threads – The number of threads used by lightgbm
early_stopping_rounds – If not None then activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training.
num_iterations – The number of boosting iterations (Default: 100)
learning_rate – Default 0.1 (constraint > 0.0) controlling the shrinkage rate
mdl_cls_obj – An optional (Default None) lightgbm model which will be used as the basis model from which training will be continued (i.e., transfer learning).
out_info_file – An output JSON file with the classification outputs scores (e.g., training, testing). If None (default) then no file is outputted.

rsgislib.classification.classlightgbm.train_opt_lightgbm_multiclass_classifier(out_mdl_file: str, cls_info_dict: Dict[str, ClassInfoObj], unbalanced: bool = False, op_mthd: int = 1, n_opt_iters: int = 100, rnd_seed: int = None, n_threads: int = 1, early_stopping_rounds: int = None, num_iterations: int = 100, max_n_leaves: int = 50, learning_rate: float = 0.1, mdl_cls_obj=None, out_params_file: str = None, out_info_file: str = None)

A function which performs a hyper-parameter optimisation for a multi-class lightgbm classifier and then trains a model saving the model for future use.

You have the option of using the bayes_opt (Default), optuna or skopt optimisation libraries. Before 5.1.0 skopt was the only option but this no longer appears to be maintained so the other options have been added.

Parameters:

out_mdl_file – The file path for the output lightgbm (*.txt) model which can be loaded to perform a classification.
cls_info_dict – a dict where the key is string with class name of ClassInfoObj objects defining the training data.
unbalanced – Boolean (Default: False) specifying whether the training data is unbalanced (i.e., a different number of samples for each class).
op_mthd – The method used to optimise the parameters. Default: rsgislib.OPT_MTHD_BAYESOPT
n_opt_iters – The number of iterations (Default 100) used for the optimisation. This parameter is ignored for skopt. For bayes_opt there is a minimum of 10 and these are added to that minimum so Default is therefore 110. For optuna this is the number of iterations used.
rnd_seed – A random seed for the optimisation. Default None. If None there a different seed will be used each time the function is run.
n_threads – The number of threads used by lightgbm
early_stopping_rounds – If not None then activates early stopping. The model will train until the validation score stops improving. Validation score needs to improve at least every early_stopping_rounds round(s) to continue training.
num_iterations – The number of boosting iterations (Default: 100)
max_n_leaves – The upper limited used within the optimisation search for the maximum number of leaves used within the model. Default: 50.
learning_rate – Default 0.1 (constraint > 0.0) controlling the shrinkage rate
mdl_cls_obj – An optional (Default None) lightgbm model which will be used as the basis model from which training will be continued (i.e., transfer learning).
out_params_file – An output JSON file with the identified parameters. If None (default) then no file is outputted.
out_info_file – An output JSON file with the classification outputs scores (e.g., training, testing). If None (default) then no file is outputted.

rsgislib.classification.classlightgbm.apply_lightgbm_multiclass_classifier(model_file: str, cls_info_dict: Dict[str, ClassInfoObj], in_msk_img: str, img_msk_val: int, img_file_info: List[ImageBandInfo], out_class_img: str, gdalformat: str = 'KEA', class_clr_names: bool = True)

A function for applying a trained multiclass lightgbm model to a image or stack of image files.

Parameters:

model_file – a trained lightgbm multiclass model which can be loaded with lgb.Booster(model_file=model_file).
cls_info_dict – a dict where the key is string with class name of ClassInfoObj objects defining the training data. This is used to define the class names and colours if class_clr_names is True.
in_msk_img – is an image file providing a mask to specify where should be classified. Simplest mask is all the valid data regions (rsgislib.imageutils.gen_valid_mask)
img_msk_val – the pixel value within the in_msk_img to limit the region to which the classification is applied. Can be used to create a hierarchical classification.
img_file_info – a list of rsgislib.imageutils.ImageBandInfo objects to identify which images and bands are to be used for the classification so it adheres to the training data.
out_class_img – The file path for the output classification image
gdalformat – The output image format (Default: KEA).
class_clr_names – default is True and therefore a colour table will the colours specified in ClassInfoObj and a class_names (from cls_info_dict) column will be added to the output file. Note the output format needs to support a raster attribute table (i.e., KEA).