NEWS.md

In bishun945/FCMm: Fuzzy Cluster Method Based on the Optimized m Value (Fuzzifier)

FCMm 0.11.1 (2021-02-22)

Critical Updates!

1. Add published Chla blending frameworks including Moore 2014, Jackson 2017, Bi 2021, and Bi thesis paper. The corresponding centroids have been attached in /inst folder. Names of these functions are Blend_Moo14, Blend_Jac17, Blend_Moo14, Blend_Bi21, and Blend_FCMm.
2. Fix bugs in plotting ggplot2 objects. As strips background use some new features by using the updated ggplot2.
3. Revise some of variable names in functions. Such as Chla_blend -> Blend_FCMm.
4. Fix bugs in some non-exported functions such as blending jobs.
5. Blending functions are moved to a new R file named Chla_blending.R. The following blending frameworks will be latterly added in this function.
6. It is a pity that in previous version the calculation form of some error metrics are in the wrong position of actual and predicted values. Anyway, the effect is only on relative metrics, for instance, some absolute metrics (like AE) is okay for that. The change in this version only switch the sign of relative metrics. Don't worry it too much!
7. In FCM-related functions, a new parameter prefix was added. So you can choose which prefix should be set before cluster number, such as "OWT", "Cluster", "Type", or whatever you like.
8. Update my old README.Rmd. :/

FCMm 0.10.3 (2021-01-22)

1. Add a new function Sampling_by_sort which returns the index of random samples by the sort.
2. Add a new param order.value to Sampling_via_cluster.

FCMm 0.10.2 (2021-01-07)

1. Add two centroids from Jackson et al. (2017) and Moore et al. (2014) with the format csv.
2. OC-serie algorithms are renamed as CHla_OCx_OLCI.
3. Add the new function OC3_OLCI.
4. Add three new functions to blend Chla estimations namely Blend_Simith, Blend_Jac17, and Blend_Moo14.

FCMm 0.10.1 (2021-01-03)

1. cal.metrics and cal.metrics.vector add new parameter c.value which is the compensated value for CAPE and CRPE series metrics. This could be useful when user wants to fix the compensated value.
2. Now the definition of valid.prediciton has been updated. The new parameter valid.definition is a list for function Assessment_via_cluster. See documents for more details.
3. I have added a new metric (temporailly) RATIO to score the candidate algorithms. So to identify the predicion with error density distribution. The center.value has been added in Assessment_via_cluster for calcluating the CAPE series metrics for RATIO.
4. A new scoring method has been added that is based on the previous sort-based method but with rescale the error metric values to avoid hidden the step values between algorithms. The function is called Scoring_algorithm_sort2 and has been coupled to Scroing_system series functions.
5. Update the calculation of Scoring_system. Now the score of accuracy and precision are restored in each of list rather the variable with XX_XXX.
6. For Scoring_system_bootstrap, I add a metric statistic for final blending result and candidates.
7. Happy new year!

FCMm 0.9.8 (2020-12-22)

1. Add a tiny example for SRF_LIST.
2. Fix a bug in apply_FCM_m about membership calculation.
3. Add a new function cal_memb to calculate distance and membership matrix but without export.

FCMm 0.9.7 (2020-12-12)

1. Add new Chla function Smith18.

FCMm 0.9.6 (2020-11-17)

1. SRF_simulate: add a parameter input_wv_as_column. Now you can directly simulate your Rrs data.frame with wavelength as column names.
2. The rgb png file output from apply_to_image has been set as false color NIR-R-G. And add a 2% linear stretch method to make the image looks more clear.

FCMm 0.9.5 (2020-11-12)

1. apply_FCM_m: the calculation of distance for large matrix has been improved.
2. SRF_simulate: add a parameter verbose to control the message to console.

FCMm 0.9.4 (2020-11-05)

Fix bugs in plot_spec_group. Parameter group_num could be inputted as characters such as "%s %s".

FCMm 0.9.3 (2020-11-02)

1. Improve the trapz function. Now better to use the trapz2 function. Much faster!
2. Add a new usefull function namely plot_spec_group which helps to plot spectra lines by means of ggplot2 with facet in different colors.

FCMm 0.9.2 (2020-11-02)

Not sure why is error report from travis so I just comment the travis lines in my readme page.

FCMm 0.9.1 (2020-11-01)

1. Parameter stand in function FuzzifierDetermination, apply_FCM_m or so has been deprecated now as users reported that it has some ambiguity about the standardized operation in FCMm. As an alternative, I set another parameter called do.stand. If do.stand == TRUE then the function will standardize the input data. The stand parameter is still be supported as before.
2. A new function (Schwammle2010) to obtain optimal m value is added, namely fDN, but it has not been exported it. Call FCMm:::fDN to run it.
3. An improved Fuzzifierdetermination version Fuzzifierdetermination2 has been created which run much faster than the former one.
4. Corresponding documents have been updated.

FCMm 0.8.7 (2020-07-10)

1. Updates documents in man.
2. Cancel the export of function read_srf_excel.
3. One of outputs by Scoring_system is changed from opt_algorithm to Opt_algorithm which is used to match with Scoring_system_bootstrap.
4. Cancel the export of functions TC2_clean and TC2_turbid since their outputs could be found in TC2_Liu20.
5. Add a new parameter pos_rgb to apply_to_image for definition of RGB channels.
6. Update README.

FCMm 0.8.6 (2020-07-01)

1. Assessment_via_cluster: update the precision calculation process which now produce NA values for extreme difference between actual and predicted values.
2. Assessment_via_cluster: for total precision calculation, it is unfair when outliers are included (which may result in extremely precision values). A quantile calculation for limiting the range of values to be calculated.
3. Assessment_via_cluster: the function will return NA values if all prediction from j algorithm in cluster i are NA values, i.e., the algorithm to be assessed is basically failed.
4. Getting_Asses_result: add a new parameter replace to control the function sample for using replacement or non-replacement.
5. Scoring_system_bootstrap: add a parameter replace to control the function sample.
6. Scoring_system_bootstrap: add an output Results_of_scoring_system to restore the all outputs from Scoring_system.
7. Scoring_system_bootstrap: fix bug of determining Opt_algorithm. Now it is determined by maximum of score values.
8. Scoring_system_bootstrap: add new item Remove_algorithm which is used for removing algorithms with huge errors when blending.
9. Scoring_system_bootstrap: update the col plotting.
10. Scoring_system_bootstrap: add Chla blending process and its corresponding plots.
11. Add a new internal function Chla_algorithm_blend which is used for blending work.
12. apply_FCM_m: fix the bug of using color_palette.
13. apply_to_image: the default setting and output control of parameters is now more reasonable.
14. Add a new function generate_param_ex, which is used for generating input for apply_to_imagewhen user-defined centroids come in.

FCMm 0.8.5 (2020-06-21)

1. Fix the bug in Assessment_via_cluster when metric values is extremely biased.
2. Fix the bug in Sampling_via_cluster when the input x has noncontinuous numeric names.
3. Fix the bug in Scoring_system when the Total_score has full NA values (for some cluster have few observation to calculate metrics.
4. Now, the parameter Times of Scoring_system_bootstrap must be larger than 1 to avoid errors.
5. New features! The Scoring_system_bootstrap now support algorithms blending by optimized candidates.
6. New features! FCM.new now support assign cluster names by sorting based on specific constrains (new parameters sort.pos and sort.decreasing).
7. New features! Added parameter color_palette for spectra plotting functions, i.e., plot_spec and apply_to_image. Users could use built-in color palettes or pre-defined color codes.
8. Documents updated.

FCMm 0.8.4 (2020-06-13)

1. Import two required packages into FCMm, i.e., scales and farver.
2. Update documents in the man folder: apply_to_image, FCM.new, Getting_Asses_results, Scoring_system.
3. I have changed the import way of the package raster since it has a conflict with base::aggregate. Thus, the raster functions are used by the way importFrom.
4. Improve the function Assessment_via_cluster. The valid percent observation is defined by values greater than zero and is.finite function.
5. Fix the bug of wrongly using SMAPE (actually is CMAPE) and so on.
6. Update the document of function Scoring_system. More and more details now! Non-related comments were removed.
7. Although the function Scoring_system could be used in the bootstrap way by users, in this version, I add a function Scoring_system_bootstrap to do that. The default sample time is set as 1000.
8. Fix the bug in function FCM.new. Previously, x.stand is x when stand = TRUE which is unfriendly for further usage. Now, both raw and normalized data are calculated in the correct way! The centroids are saved on both raw and normalized scales.
9. Add the function HUE which is the default color palette of the package ggplot2.
10. Delete rm(list=ls()) in Builtin_centroids.Rmd.
11. Update documents in Cluster_new_data.Rmd.
12. Update README files.

FCMm 0.8.3 (2020-06-12)

1. Thanks for the bug report by Xiaolan Cai, a new return centroids of function FCM.new is supported now, which indicates the cluster centroids of your training data set.
2. Add a new ifnotstop for function Assessment_via_cluster that ZERO values in measured vector will introduce Inf metric results for cal.metrics or cal.metrics.vector.
3. Add a new color palette function scales::hue_pal which is the default palette of package ggplot2.

FCMm 0.8.2 (2020-06-11)

1. This update is for revision from CRAN manual inspection.
2. All examples in Rd files were unwrapped which used \dontrun{} except for apply_to_image.Rd as its executable time is > 5 sec.
3. Functions cal.metrics.names, cal.metircs.vector.names, Chla_algorithms_name, and show_sensor_names now are alias of cal.metrics, cal.metrics.vector, run_all_Chla_algorithms, and SRF_simulate respectively.
4. In this version, on.exit is used in the right position which is next to oldoptions <- options(scipen=1000) #247 of Image_application.R.
5. The LICENSE and cph are fixed in this version.
6. Updated the function plot_spec. For convenience of plotting spectra by groups. Now the facet_wrap are used rather than restored in a list. Therefore, its argument HABc is deprecated in this version.
7. Documents are updated for grammar and spelling issues.
8. TSM estimation algorithm GAA_SPM by Xiaolong Yu is supported in this version.

FCMm 0.8.1 (2020-06-06)

1. This is a major update for FCMm.
2. Updated DESCRIPTION by adding fields BugReports and Authors@R for more meta information. The previous Author and Maintainer were deleted.
3. Rd files in folder man were updated including apply_FCM_m, apply_to_image, Assessment_via_cluster, Bloom, BR_Gil10, BR_Git11, C6, cal.metrics.names, cal.metrics, cal.metrics.vector.names, cal.metrics.vector, Chla_algorithms_name, FBA_Le13, FBA_Yang10, FCM.new, FuzzifierDetermination, generate_param, Getting_Asses_results, Gons08, level_to_variable, lmodel2_, NDCI_Mi12, OC4_OLCI, OC5_OLCI, OC6_OLCI, plot_spec, plot_spec_from_df, QAA_v5, RdYlBu, read_srf_excel, run_all_Chla_algorithms, Sampling_via_cluster, SCI_Shen10, Score_algorithm_interval, Score_algorithm_sort, Scoring_system, show_sensor_names, Spectral, SRF_simulate, TBA_Gil10, TBA_Git11, TC2, TC2_clean, TC2_turbid, trapz, and trim_sd.
4. Updated README files by adding a new chunk for the assessment part.
5. Updated NAMESPACE.
6. All logical variables T and F were changed by TRUE and FALSE.
7. There were naming mistakes in tools.R for metrics functions cal.metrics and cal.metrics.vector which replace the SAPE (symmetric) series with the CAPE (compensated) series.
8. Updated run_all_Chla_algorithms which could run without constrains for wavelength settings as algorithms with missing wavelength will return NA values alternatively.
9. Renamed and updated vignettes for simplification Assessment (new), Builtin_centrodis, and Cluster_new_data.
10. Codes about options() were removed or used by on.exit().

FCMm 0.7.5 (2020-06-02)

# As the version 0.7.4 was submitted to CRAN, I updated DESCRIPTION file follow the requirement by the CRAN which also includes LICENSE updated by using uesthis::use_mit_license(name="Shun Bi").

FCMm 0.7.4 (2020-06-02)

Found and fixed an Unicode \u2010 in one Rd files of man which may result in the error check by rhub.

FCMm 0.7.3 (2020-06-01)

1. Refresh the Rd files in man.
2. Export the function trapz().
3. Updated README and vignettes.

FCMm 0.7.2 (2020-06-01)

1. Deleted the suggested package dplyr as I can use base::sample as an alternative.
2. The previous four vignettes are shorten to the two that are Cluster_a_new_dataset_by_FCM and Usage_of_the_built_in_centroids_by_FCMm which might provide enough help for users.
3. Files in DOCS are updated.
4. README updated.

FCMm 0.7.1 (2020-05-30)

1. Updated README files of which superlinks for vignettes are deleted as the nexting version will include other vignettes or some modifications.
2. Renamed the filename of vignettes, also, of which superlinks are deleted. Several old version vignettes on DOC folder are also deleted.
3. Fixed spelling errors in Rd files.
4. In this version, the imported package - heatmaply - was deleted as it would include warning information when library(FCMm). Anyway, the functions supported by heatmaply, i.e., Spectral() and RdYlBu() are imported from packages grDevices::colorRampPalette(RColorBrewer::brewer.pal(11, "Spectral") and ... ,"RdYlBu").
5. Added cran-comments.md file as I decide to release this package to CRAN.
6. Updated man Rd files.

FCMm 0.6.2 (2020-05-24)

• Bugs fixed.

FCMm 0.6.1 (2020-05-23)

• This version updated the arrangement of codes, fixed bugs, and collected data sets.
• Ignored files are removed from Gihub.
• I gonna resolve the problem of do parameters in next versions.

FCMm 0.5.7 (2020-05-12)

• Bugs of SRF_simulate fixed.

FCMm 0.5.6 (2020-03-26)

• Bugs of apply_to_image fixed.

FCMm 0.5.5 (2020-03-17)

• Hey guys, due to the outbreak of coronavirus, I have stopped the update for a long time.
• In this version. The assessment method in FCMm has been improved by a new proposed scoring rule called sort-based method.
• I am collecting the data and references about the algorithm assessment and finally going to finish my manuscript. Hope this will help to improve the algorithm evaluation work the in water color community.
• Please, COVID-19 go away!

FCMm 0.5.4 (2020-03-03)

• Update the document of data OLCI_TH.
• Improve the effectiveness of function apply_FCM_m as the matrix may speed up the process rather than the data.frame (I suppose).

FCMm 0.5.3 (2020-02-28)

• Add newly trained Bloom model which was used to replace C6 model.

FCMm 0.5.2 (2020-02-22)

• Bugs of Sampling_via_cluster were fixed.

FCMm 0.5.1 (2020-02-21)

• Create the new file QAA_series.R to add functions TC and QAA_v5. The original functions from FCM_m_Chla_estimation were removed accordingly.
• Create the new file Algorithms_assessment.R to add functions Assessment_via_cluster, Score_algorithms, Sampling_via_cluster, Getting_Asses_result, and Scoring_system. The original function Assessment_via_cluster was removed from Algorithms_assessment.R.
• See more details in help documents of these new functions.

FCMm 0.4.17 (2020-02-20)

• Modify the constrain of SMA linear regression such as Rsquare value and point number taken part into the calculation.
• Next update is a first-point caning version, 0.5.X since the new assessment system (scoring system) will be load ASAP.

FCMm 0.4.16 (2020-02-19)

• Add several new Error Metrics into tools.R.
• Use Type-II regression, namely SMA (Standard Major Axis), to obtain the regression results, such as Intercept and Slope. The main reason for that is, inputs of regression (for validation) are obtained without fully quality control (still have uncertainties in each variable). This kind of regression, comparing the Type-I, cannot be used for prediction (such as using Rrs to predict Chla concentrations), but better to compare two variables on the same scale (units), useful for the error assessment.

FCMm 0.4.15 (2020-02-17)

• Add linear regression results (slope and intercept) to tool.R for hard mode assessment.

FCMm 0.4.14 (2020-02-16)

• Add new plot option for Assessment_via_cluster

FCMm 0.4.13 (2020-02-15)

• Result Area was added in the return of apply_FCM_m.

FCMm 0.4.12 (2020-02-14)

• Bugs of TC2 fixed.

FCMm 0.4.11 (2020-02-13)

• Bugs of BR_Gil10 and NDCI_Mi12 fixed.

FCMm 0.4.10 (2020-02-13)

• Bugs of FCM_m_Chla_estimation fixed.

FCMm 0.4.9 (2020-02-10)

• Add UMRPE function to calculate unbiased error metric.

FCMm 0.4.8 (2020-02-09)

• Update the function cal.metrics which remove the double version of .cal.mae or .bias. The log10-transformation is now supported by the logical parameter log10.
• Add functions cal.metrics.vector and cal.metrics.vector.names which calculate the validation metrics via pairwise way (results are columns).
• Update the function Assessment_via_cluster by adding a fuzzy mode to calculate validation metrics via membership values.

FCMm 0.4.7 (2020-02-08)

• Add QAA_v5 in file FCM_m_Chla_estiamtion.
• Some bugs fixed.

FCMm 0.4.6 (2020-02-05)

Update the functions and references in file FCM_m_Chla_estiamtion.

Notes (2020-01-21)

• Even though the coming phylogenetic tree method will offer a good option to post-classify the numerous clusters, we still have to design a better allocation strategy on cluster number. That's to say, the nearby cluster name (for instance, Cluster 1 and Cluster 2) should have the similar optical attribution. However, the current FCM.new or FuzzifierDetermiantion functions return the totally random results. Should be improved later!
• Another confusing thing is the parameter stand in many functions (for instance, FCM.new). The original design of stand is making sure the input x has been standardized so that the median-process of clustering should not standardize that. Actually, the parameter could be better understood if named as DoStand. When DoStand is True, it means the input x have to be standardized in the following process. This change will make numerous source codes different so I will get to do it after other functions are gradually improved.

FCMm 0.4.5 (2020-01-21)

• Fix bugs of function Assessment_via_cluster to avoid produce NA values.

FCMm 0.4.4 (2020-01-20)

• Add new function run_all_Chla_algorithms to obtain the Chla concentration if required Rrs bands are get.
• Group the function list by adding the @family tag using package roxygen2.

FCMm 0.4.3 (2020-01-19)

• Supply the help documents of function SRF_simulate.
• Add an option of function SRF_simulate --- wv_as_column which makes the output as a dataframe with wavelength as column name. Easy to be an input of function apply_FCM_m or so.
• In next version, I gonna add the Chla estimation method proposed by Liu et al. (2020) which has been accepted by RSE.

FCMm 0.4.1 (2020-01-12)

• Recently, I reread Moore's and Jackson's papers and have some new insights.

  • I should re-consider the FCM process when input data (here is Rrs spectra) is biased.
  • Mahalanobis distance should be under consideration since it eliminate the influence of variable correlation. However, it is questionable to define the covariance when facing new input data.
  • Moore et al. (2001) supposes the distribution of Rrs vectors belonging to one class is multivariate normal, and Rrs belongs to that class, the distance metric has a chi-sq distribution with n degrees of free (where n is the dimension of Rrs). Is this evidential in current data sets?

• Phylogenetic tree is an effective method to post-classification (or post-group) of initial cluster results.

• This mode will be added in version 0.5.X or later (2020-01-19).

• How to assign new data to an existing clustering?

• Still consider this problem ... If you guys have some new ideals, please let me know.

FCMm 0.4.0 (2020-01-10)

• The input of apply_FCM_m() with built-in clusters should match OLCI band settings.
• However, the data of users may be hyperspectral Rrs. Need a function that convert hyperspectral Rrs into the multispectral.
• Should add a function like SRF_simulate() to simulate the Rrs by spectral response function.

• Meanwhile, the SRF_list.rda as a list with SRF data of many sensors should be added.

• Need another vignette to illustrate how we can bootstrap the training set for optimizing the cluster number or fuzzifier value.

• Now the result of function apply_FCM_m() is plotted on single picture which is hard to read when the number of spectra is large. So I gonna add a new plot function by using facet method from ggplot2.

• A new result of apply_FCM_m() is now added as p.group.facet. You could yourresult$p.group.facet to obtain the spectra with facet.

FCMm 0.3.3 (2020-01-09)

• Add NEWS.md file in the top-level fold so that you could see what changes in each version of FCMm.

bishun945/FCMm documentation built on Oct. 15, 2021, 6:43 p.m.