Performance analyses of classifier ensembles on peptide encodings

public public 1yr ago 0 bookmarks
View Workflow
Report
ensemble-performance
ensemble-performance
ensemble-performance
View Workflow

image info

Overview image illustrating the performance analyses of classifier ensembles on peptide encodings. An arbitrary number of encoded datasets can be processed (arrows). The workflow conducts preprocessing (a), ensemble pruning (b), training/testing (c), and processing of results (d) using a 100-fold Monte Carlo cross-validation. Manuscript submitted for publication.

Execution

To run all experiments, execute snakemake --cores 32 --quiet .

In order to execute the ensemble pruning using the Decision Tree classifier, stacking as the meta-model, with 5 folds, run snakemake data/temp/avp_amppred/ensemble_pfront/stacking/dt/{0,1,2,3,4}.csv --cores 32 --quiet

This example will run the pruning for the avp_amppred dataset.

Code Snippets

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Sys.setlocale("LC_NUMERIC","en_US.UTF-8")

library(scmamp)
library(yaml)

paths <- snakemake@input

d <- read.csv(paths[[1]])
d["X"] <- NULL
for (idx in 2:length(paths)) {
  df = read.csv(paths[[idx]])
  df["X"] <- NULL
  d <- rbind(d, df)
}

d <- d[(d$cat != "mvo"), ]

for (c in unique(d[["cat"]])) {
  d[d["cat"] == c, "idx"] <- 1:sum(d["cat"] == c)
}

df_cats <- reshape(d[, -2:-4], idvar = "idx", timevar = "cat", direction = "wide")
df_cats$idx <- NULL
colnames(df_cats) <- gsub("mcc.", "", colnames(df_cats))

# plotCD(results.matrix = df_cats, alpha = 0.05)

ar_cats <- sort(colMeans(rankMatrix(df_cats)))

nm_cats <- nemenyiTest(df_cats)
rownames(nm_cats$diff.matrix) <- colnames(nm_cats$diff.matrix)

lists <- apply(
  X = expand.grid(unique(d[["model"]]), unique(d[["meta_model"]])),
  MARGIN = 1,
  FUN = function(row) {
    df <- data.frame(d[d["model"] == row[1] & d["meta_model"] == row[2], "mcc"])
    colnames(df) <- paste0(row[1], "_", row[2])
    return(df)
  }
)

df_models <- do.call(cbind, lists)

ar_models <- sort(colMeans(rankMatrix(df_models)))

nm_models <- nemenyiTest(df_models)
rownames(nm_models$diff.matrix) <- colnames(nm_models$diff.matrix)

lres <- list(
  cats = list(
    cd = nm_cats$statistic,
    average_ranking = ar_cats,
    names = names(ar_cats)
  ),
  models = list(
    cd = nm_models$statistic,
    average_ranking = ar_models,
    names = names(ar_models)
  )
)

write_yaml(lres, snakemake@output[[1]])
R yaml From line 1 of scripts/cd.R 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
import pandas as pd
import altair as alt
import numpy as np

df_res = pd.read_csv(snakemake.input[0], index_col=0)

df_out = pd.DataFrame()
for m in df_res.model.unique():
    df_tmp = df_res.loc[df_res.model == m]
    df_tmp = df_tmp.loc[np.bitwise_not(
        df_tmp.ensemble_mvo |
        df_tmp.ensemble_best |
        df_tmp.ensemble_rand |
        df_tmp.ensemble_chull |
        df_tmp.ensemble_pfront
    ) & (df_tmp.chull_complete == -1)]
    df_tmp = pd.concat([
        pd.DataFrame({"variable": df_tmp.x, "type": "kappa", "model": m}),
        pd.DataFrame({"variable": df_tmp.y, "type": "error", "model": m})
    ])
    df_out = pd.concat([df_out, df_tmp])

chart = alt.Chart(df_out).mark_boxplot(
    color="grey",
    size=15
).encode(
    x=alt.X("type:N", title=None, axis=None),
    y=alt.Y("variable:Q", title=None),
    color=alt.Color(
        "type:N", title="Type",
        scale=alt.Scale(scheme="greys")
    ),
    column=alt.Column("model:N", title="Model", spacing=2)
).properties(
    width=50,
    height=100
)

chart.save(snakemake.output[0])  # html
Python Pandas numpy altair From line 1 of plots/box_plot_manova.py 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
import pandas as pd
import altair as alt

from glob import glob

df_res = pd.DataFrame()
# for p in glob("data/ensembles_res/*/*/*.csv"):
for p in list(snakemake.input.ensemble_res):
    df_tmp = pd.read_csv(p, index_col=0)
    df_res = pd.concat([df_res, df_tmp])

df_res_single = pd.DataFrame()
for p in list(snakemake.input.single_res):
    df_tmp = pd.read_csv(p, index_col=0)
    df_res_single = pd.concat([df_res_single, df_tmp])

c1 = alt.Chart(df_res).mark_boxplot(
    size=8, color="#000000", opacity=1.0, outliers={"size": 0}, median=False
).encode(
    x=alt.X("meta_model:N", title=None, axis=alt.Axis(labelAngle=-35, grid=True)),
    y=alt.Y(
        "mcc:Q",
        scale=alt.Scale(domain=[0.0, 1.0]),
        axis=alt.Axis(values=[0.1, 0.3, 0.5, 0.7, 0.9], title=None)
    ),
).properties(
    width=100,
    height=100
).facet(
    row=alt.Row("model:N", title=None),
    column=alt.Column("cat:N", title=None, sort=["pfront", "chull", "mvo", "best", "rand"]),
    spacing=1
)

c2 = alt.Chart(df_res_single).mark_boxplot(
    size=8, color="#000000", opacity=1.0, outliers={"size": 0}, median=False
).encode(
    x=alt.X("rank:N", title=None, axis=alt.Axis(labelAngle=-35, grid=True)),
    y=alt.Y(
        "mcc:Q",
        scale=alt.Scale(domain=[0.0, 1.0]),
        axis=alt.Axis(values=[0.1, 0.3, 0.5, 0.7, 0.9], title=None, orient="right")
    ),
).properties(
    width=100,
    height=100
).facet(
    row=alt.Row("model:N", title=None, header=alt.Header(title=None, labels=False)),
    column=alt.Column("cat:N", title=None, sort=["pfront", "chull", "mvo", "best", "rand", "single_best"]),
    spacing=1
)

chart = alt.hconcat(
    c1, c2, spacing=0.7
).configure_header(
    labelFontSize=14
).configure_axis(
    labelFontSize=12
)

chart.save(snakemake.output[0], vegalite_version="5.1.0")  # html
chart.save(snakemake.output[1], vegalite_version="5.1.0")  # png
Python Pandas altair From line 1 of plots/box_plot.py 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
from more_itertools import chunked

import altair as alt
import pandas as pd

res2 = []
for p in list(snakemake.input):
    mmodel = p.split("/")[4]
    model = p.split("/")[5]
    fold = int(p[-5:-4])
    with open(p) as f:
        res = list(chunked(f.readlines(),6))
        for idx, l in enumerate(res):
            fitness, mcc = l[2].rstrip().split(",")
            fitness = float(fitness.replace("Best Fitness: ",""))
            mcc = float(mcc.replace(" best metrics: {'mcc': ","").replace("}",""))
            res2.append([idx, fitness, mcc, fold, model, mmodel])

source = pd.DataFrame(res2,columns=["gen", "fitness", "mcc", "fold", "model", "mmodel"])

line = alt.Chart(source).mark_line(color="black").encode(
    x="gen:O",
    y="mean(fitness):Q"
)

band = alt.Chart(source).mark_errorband(extent="ci", color="black").encode(
    x=alt.X("gen:O", title=None, axis=alt.Axis(labelAngle=-35)),
    y=alt.Y("fitness:Q", title=None)
)

chart = (band + line).properties(
    width=100,
    height=100
).facet(
    column=alt.Column("model:N", title=None),
    row=alt.Row("mmodel:N", title=None),
    spacing=1
).configure_header(
    labelFontSize=14
).configure_axis(
    labelFontSize=12
)

chart.save(snakemake.output[0])  # html
chart.save(snakemake.output[1])  # png
Python Pandas altair more-itertools From line 1 of plots/gens_vs_perf.py 
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
import pandas as pd
import altair as alt
import numpy as np

df_res = pd.read_csv(snakemake.input[0], index_col=0)
# df_res = pd.read_csv("data/temp/amp_antibp2/kappa_error_res/plot_data.csv", index_col=0)

x_min, y_min = df_res.loc[df_res.fold == 0].x.min(), df_res.loc[df_res.fold == 0].y.min()
x_max, y_max = df_res.loc[df_res.fold == 0].x.max(), df_res.loc[df_res.fold == 0].y.max()

df_res = df_res.loc[(df_res.x >= x_min) & (df_res.x <= x_max) & (df_res.y >= y_min) & (df_res.y <= y_max)]

# adopted from https://realpython.com/python-rounding/#rounding-up
def round_up(n, decimals=0):
    multiplier = 10 ** decimals
    return np.ceil(n * multiplier) / multiplier

# adopted from https://realpython.com/python-rounding/#rounding-down
def round_down(n, decimals=0):
    multiplier = 10 ** decimals
    return np.floor(n * multiplier) / multiplier


x_min = round_down(x_min, 1)
y_min = round_down(y_min, 1)

scatter = alt.Chart().mark_point(filled=True, opacity=1.0).encode(
    x=alt.X(
        "x:Q", title="kappa",
        scale=alt.Scale(domain=[x_min, x_max])
    ),
    y=alt.Y(
        "y:Q", title="average pair-wise error", axis=alt.Axis(grid=True),
        scale=alt.Scale(domain=[y_min, y_max])
    ),
    color=alt.Color(
        "cat:N", title="Pruning",
        scale=alt.Scale(
            domain=["all", "best", "chull", "mvo", "pfront", "rand"],
            range=["gray", "#fdae61", "#2c7bb6", "yellow", "#d7191c", "#abd9e9"]),
        legend=alt.Legend(orient="bottom", offset=12)
    ),
    size=alt.condition(
        alt.datum.cat == "all",
        alt.value(50),
        alt.value(100)
    ),
).properties(
    width=300,
    height=200
)

convex_hull = alt.Chart().mark_line(
    color="#2c7bb6",
    size=1.1
).encode(
    x=alt.X("x:Q", title=None),
    y=alt.Y("y:Q", title=None),
    order="chull:N",
).transform_filter(
    alt.datum.chull != -1
)

pareto_frontier = alt.Chart().mark_line(
    strokeDash=[5, 1],
    color="#d7191c",
    size=1.1
).encode(
    x="x:Q",
    y="y:Q",
    order="pfront:N"
).transform_filter(
    alt.datum.pfront != -1
)

vals = np.array(range(51)) / 100
vals = [e for e in vals if e <= y_max]
df = pd.DataFrame({"x": [1 - (1 / (1 - i)) for i in vals], "y": vals})
df = df.loc[(df.x >= x_min) & (df.y >= y_min)]

bound_line = alt.Chart(df).mark_line(color="gray", strokeDash=[4, 4]).encode(
    x=alt.X("x:Q"),
    y="y:Q"
)

c1 = alt.layer(
    convex_hull,
    pareto_frontier,
    scatter,
    bound_line,
    data=df_res.loc[df_res.fold == 0]
).facet(
    row=alt.Column("model", title=None),
    spacing=10
)

heatmap = alt.Chart().mark_rect().encode(
    x=alt.X(
        "x:Q",
        title=None,
        bin=alt.Bin(maxbins=40),
        axis=alt.Axis(values=[-1.0, -0.5, 0.0, 0.5, 1.0], format=".1f", grid=True),
        scale=alt.Scale(domain=[x_min, x_max])
    ),
    y=alt.Y(
        "y:Q",
        title=None,
        bin=alt.Bin(maxbins=40),
        axis=alt.Axis(
            values=[0.0, 0.1, 0.2, 0.3, 0.4, 0.5], 
            format=".1f", 
            grid=True, domain=False, 
            ticks=False, labels=False
        ),
        scale=alt.Scale(domain=[y_min, y_max])
    ),
    color=alt.Color(
        "count(x):Q",
        title="Count",
        legend=alt.Legend(
            gradientLength=90,
            orient="bottom",
            offset=12
            # values=[0, 1500, 3000, 4500]
            # values=[0, np.histogram2d(x=df_res.x, y=df_res.y, bins=45)[0].max()]
        ),
        scale=alt.Scale(scheme="greys")
    ),
    tooltip="count(x):Q"
).properties(
    width=300,
    height=200
)

c2 = alt.layer(
    heatmap,
    bound_line,
    # data=df_res.loc[df_res.fold.isin([0, 1, 2, 3, 4, 5])].reset_index()
    data=df_res.reset_index()
).facet(
    row=alt.Row(
        "model:N",
        title=None,
        header=alt.Header(labels=False)
    ),
    spacing=10
)

resc = alt.hconcat(
    c1,
    c2,
    spacing=1
).resolve_scale(
    color="shared"
)

chart = resc.configure_header(
    labelFontSize=14
).configure_axis(
    labelFontSize=12
).configure_legend(
    gradientThickness=10,
    labelFontSize=13,
    columns=3
)

chart.save(snakemake.output[0])  # html
Python Pandas numpy altair From line 1 of plots/kappa_error.py 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
from xcd_plot import XCDChart

import pandas as pd
import altair as alt

import yaml

# with open("data/ensembles_res/cd.yaml") as f:
with open(snakemake.input[0]) as f:
    cd_data = yaml.safe_load(f)

from glob import glob

df_res = pd.DataFrame()
# for p in glob("data/ensembles_res/*/*/*.csv"):
for p in list(snakemake.input)[1:]:
    df_tmp = pd.read_csv(p, index_col=0)
    df_res = pd.concat([df_res, df_tmp])

df_res = df_res.loc[df_res.cat != "mvo"]

from pprint import pprint

s = sorted(zip(cd_data["models"]["average_ranking"], cd_data["models"]["names"]), key=lambda tup: tup[1])

# cd_data["models"]["average_ranking"] = [e[0] for e in s]
# cd_data["models"]["names"] = [e[1] for e in s]

xcd_chart = XCDChart(ensemble_data=df_res, cd_data=cd_data)
xcd_chart.save(snakemake.output[0])  # html
xcd_chart.save(snakemake.output[1])  # png
Python Pandas PyYAML altair From line 1 of plots/xcd.py 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
library(broom)

d <- read.csv(snakemake@input[[1]])
# d <- read.csv("data/temp/avp_amppred/kappa_error_res/plot_data.csv")

res.man <- manova(cbind(d$x, d$y) ~ model, data = d)

manova_summary <- summary(res.man)
write.csv(tidy(res.man), snakemake@output[[1]])

manova_summary_aov <- summary.aov(res.man)
write.csv(rbind(
  data.frame(
    manova_summary_aov[[1]],
    response=names(manova_summary_aov[1])
  ),
  data.frame(
    manova_summary_aov[[2]],
    response=names(manova_summary_aov[2])
  )
), snakemake@output[[2]])

df_res <- do.call(rbind, lapply(unique(d[["model"]]), function(m) {
  d_tmp <- d[d["model"] == m, ]
  d_tmp <- d_tmp[
    (d_tmp$ensemble_best == "False") &
    (d_tmp$ensemble_rand == "False") &
    (d_tmp$ensemble_chull == "False") &
    (d_tmp$ensemble_pfront == "False") &
    (d_tmp$chull_complete == -1)
  , ]
  # d_tmp <- d_tmp[sample(nrow(d_tmp), 1000), ]
  data.frame(kappa = d_tmp[ ,"x"], error = d_tmp[ ,"y"], model = m)
}))

anova_kappa_aov <- aov(df_res$kappa ~ df_res$model)
write.csv(tidy(anova_kappa_aov), snakemake@output[[3]])

anova_kappa_tukey_hsd <- TukeyHSD(aov(df_res$kappa ~ df_res$model))
write.csv(tidy(anova_kappa_tukey_hsd), snakemake@output[[4]])

anova_error_aov <- aov(df_res$error ~ df_res$model)
write.csv(tidy(anova_error_aov), snakemake@output[[5]])

anova_error_tukey_hsd <- TukeyHSD(aov(df_res$error ~ df_res$model))
write.csv(tidy(anova_error_tukey_hsd), snakemake@output[[6]])

### areas

df_res <- do.call(
  rbind,
  lapply(
    snakemake@input[2:5],
    # Sys.glob("data/temp/avp_amppred/areas/*/res.csv"),
    read.csv
  )
)

anova_area_aov <- aov(df_res$area ~ df_res$model)
write.csv(tidy(anova_area_aov), snakemake@output[[7]])

anova_area_tukey_hsd <- TukeyHSD(aov(df_res$area ~ df_res$model))
write.csv(tidy(anova_area_tukey_hsd), snakemake@output[[8]])
R broom From line 1 of scripts/statistics.R 
106
107
108
109
110
111
112
113
114
115
116
117
118
run:
    dict_indcs = {}
    indcs = []
    for p in list(input):
        df = pd.read_csv(p, index_col=0)
        dict_indcs = dict_indcs | dict(df["y"])
        indcs += [list(df.index)]

    df_res = pd.DataFrame(dict_indcs.items(), columns=["idx", "y"])

    # get common all indices
    indcs = sorted(functools.reduce(set.intersection, indcs[1:], set(indcs[0])))
    df_res.loc[df_res.idx.isin(indcs)].to_csv(output[0])
SnakeMake From line 106 of main/Snakefile 
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
run:
    df_indices = pd.read_csv(input[0], index_col=0)
    indices, y = df_indices["idx"].values, df_indices.y.values

    gss = StratifiedShuffleSplit(n_splits=len(FOLDS), train_size=.8, random_state=42)

    df_train, df_val, df_test = pd.DataFrame(), pd.DataFrame(), pd.DataFrame()
    for train_idx, test_idx in gss.split(indices, y):
        val_indcs = test_idx[:math.ceil(len(test_idx) / 2)]
        test_indcs = test_idx[math.ceil(len(test_idx) / 2):]
        ser_indices_train = df_indices.iloc[train_idx, :]["idx"]
        df_train = pd.concat([
            df_train,
            ser_indices_train.reset_index(drop=True)
        ], axis=1)
        ser_indices_val = df_indices.iloc[val_indcs, :]["idx"]
        df_val = pd.concat([
            df_val,
            ser_indices_val.reset_index(drop=True)
        ],axis=1)
        ser_indices_test = df_indices.iloc[test_indcs, :]["idx"]
        df_test = pd.concat([
            df_test,
            ser_indices_test.reset_index(drop=True)
        ], axis=1)

    df_train.columns = [f"fold_{i}" for i in FOLDS]
    df_val.columns = [f"fold_{i}" for i in FOLDS]
    df_test.columns = [f"fold_{i}" for i in FOLDS]

    df_train.to_csv(output[0])
    df_val.to_csv(output[1])
    df_test.to_csv(output[2])
SnakeMake From line 128 of main/Snakefile 
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
run:
    df_indices = pd.read_csv(input[0], index_col=0)
    indices = df_indices["idx"].values

    df = pd.read_csv(input[1], index_col=0).loc[indices, ]
    X, y = df.iloc[:, :-1].values, df["y"].values
    X_scaled = MinMaxScaler().fit_transform(X)

    vals = np.hstack((X_scaled, y.reshape((y.shape[0], 1))))

    indices = np.argwhere(pd.DataFrame(vals).std().values == 0).flatten()
    vals = np.delete(vals, indices, 1)

    if len(indices) != 0:
        print(wildcards.csv_name)

    df_res = pd.DataFrame(vals, columns=np.delete(df.columns, indices), index=df.index)
    df_res.to_csv(output[0])
SnakeMake From line 169 of main/Snakefile 
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
run:
    df_indcs_train = pd.read_csv(input[1], index_col=0)
    indcs_train_tmp = df_indcs_train[f"fold_{wildcards.fold}"]
    df_indcs_val = pd.read_csv(input[2],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]
    indcs_train = pd.concat([indcs_train_tmp, indcs_val])

    df_indcs_test = pd.read_csv(input[3], index_col=0)
    indcs_test = df_indcs_test[f"fold_{wildcards.fold}"]

    df = pd.read_csv(input[0], index_col=0)
    X_train = df.iloc[:, :-1].loc[indcs_train, :].values
    y_train = df.loc[indcs_train, "y"].values
    X_test = df.iloc[:, :-1].loc[indcs_test, :].values
    y_test = df.loc[indcs_test, "y"].values

    clf = MODEL[wildcards.model]
    try:
        clf.fit(X_train, y_train)
        y_pred = clf.predict(X_test)
        mcc = matthews_corrcoef(y_test, y_pred)
    except np.linalg.LinAlgError as e:
        print(e)
        mcc = 0.0
    except ValueError as e:
        print(e)
        mcc = 0.0

    pd.DataFrame({
        "mcc": [mcc],
        "fold": [wildcards.fold],
        "encoding": [wildcards.csv_name],
        "model": [wildcards.model]
    }).to_csv(output[0])
SnakeMake From line 196 of main/Snakefile 
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
run:
    df_res = pd.DataFrame()
    for p in list(input):
        df_tmp = pd.read_csv(p, index_col=0)
        df_res = pd.concat([df_res, df_tmp])

    res = df_res.groupby("encoding")\
        .apply(lambda df: df.mcc.mean())\
        .sort_values(ascending=False)

    top3 = res.index[:3].to_list()

    df_res = df_res.loc[df_res.encoding.isin(top3)]
    df_res["rank"] = -1
    df_res["cat"] = "single"

    for i, enc in enumerate(top3, start=1):
        df_res.loc[df_res.encoding == enc, "rank"] = f"Top_{i}"

    df_res.to_csv(output[0])
SnakeMake From line 238 of main/Snakefile 
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
run:
    df_indcs_train = pd.read_csv(input[0], index_col=0)
    indcs_train = df_indcs_train[f"fold_{wildcards.fold}"]

    df_indcs_val = pd.read_csv(input[1],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]

    paths = list(input[2:])

    encoded_datasets = [pd.read_csv(p, index_col=0) for p in paths]

    X_train_list = \
        [df.loc[indcs_train, :].iloc[:, :-1].values
         for df in encoded_datasets]

    X_val_list = \
        [df.loc[indcs_val, :].iloc[:, :-1].values
         for df in encoded_datasets]


    y_train, y_val = \
        encoded_datasets[0].loc[indcs_train, "y"].values, \
        encoded_datasets[0].loc[indcs_val, "y"].values

    clf = MODEL[wildcards.model]
    eclf = META_MODEL[META_MODELS[0]]
    eclf.estimators = [(paths[i], clf) for i in range(len(paths))]
    eclf.fit(X_train_list, y_train)

    res = []
    for ((e1, clf_1), X_val_1), ((e2, clf_2), X_val_2) in \
            itertools.combinations(zip(eclf.estimators_, X_val_list), 2):
        y_pred_tree_1, y_pred_tree_2 = \
            clf_1.predict(X_val_1), clf_2.predict(X_val_2)
        error_1, error_2 = \
            1 - accuracy_score(y_pred_tree_1, y_val), \
            1 - accuracy_score(y_pred_tree_2, y_val)
        mean_pairwise_error = np.mean([error_1, error_2])
        k = kappa(y_pred_tree_1,y_pred_tree_2)
        res += [[k, mean_pairwise_error, e1, e2]]

    df_res = pd.DataFrame(res, columns=["x", "y", "encoding_1", "encoding_2"])
    df_res["model"] = wildcards.model

    df_res.to_csv(output[0])
SnakeMake From line 291 of main/Snakefile 
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
run:
    df_points = pd.read_csv(input[0], index_col=0)

    hull = ConvexHull(df_points[["x", "y"]])

    df_points["chull_complete"] = -1
    df_points.iloc[hull.vertices, df_points.columns.get_loc("chull_complete")] = \
        range(hull.vertices.shape[0])

    df_points.to_csv(output[0])

    pd.DataFrame({
        "model": [wildcards.model],
        "area": [hull.area],
        "fold": [wildcards.fold]
    }).to_csv(output[1])
SnakeMake From line 343 of main/Snakefile 
367
368
369
370
371
372
run:
    df_res = pd.DataFrame()
    for p in list(input):
        df_res = pd.concat([df_res, pd.read_csv(p, index_col=0)])

    df_res.to_csv(output[0])
SnakeMake From line 367 of main/Snakefile 
407
408
409
410
411
412
413
414
415
416
417
418
419
420
run:
    df_points = pd.read_csv(input[0], index_col=0)

    df_hull = df_points.loc[df_points.chull_complete != -1, ["x", "y"]]

    # mask convex hull (use only vals towards lower, left corner)
    P = pareto_n(-df_hull.values)

    indices = list(df_hull.iloc[P[0], :].sort_values("x").index)

    df_points["chull"] = -1
    df_points.iloc[indices, df_points.columns.get_loc("chull")] = range(len(indices))

    df_points.to_csv(output[0])
SnakeMake From line 407 of main/Snakefile 
427
428
429
430
431
432
433
434
435
436
437
run:
    df_points = pd.read_csv(input[0], index_col=0)

    P = pareto_n(-df_points[["x", "y"]].values)

    indices = list(df_points.iloc[P[0], :].sort_values("x").index)

    df_points["pfront"] = -1
    df_points.iloc[indices, df_points.columns.get_loc("pfront")] = range(len(indices))

    df_points.to_csv(output[0])
SnakeMake From line 427 of main/Snakefile 
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
run:
    df_indcs_train = pd.read_csv(input[0],index_col=0)
    indcs_train_tmp = df_indcs_train[f"fold_{wildcards.fold}"]
    df_indcs_val = pd.read_csv(input[1],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]
    indcs_train = pd.concat([indcs_train_tmp, indcs_val])

    df_indcs_test = pd.read_csv(input[2],index_col=0)
    indcs_test = df_indcs_test[f"fold_{wildcards.fold}"]

    df_points = pd.read_csv(input[3], index_col=0)

    # y is average pairwise error
    train_paths = list(set(
        df_points\
            .sort_values("y").iloc[:15, :][["encoding_1", "encoding_2"]]\
            .values.flatten()
    ))

    # keep ensemble best encodings position for later usage
    indices = df_points.sort_values("y").iloc[:15, :].index
    df_points["ensemble_best"] = False
    df_points.iloc[indices, df_points.columns.get_loc("ensemble_best")] = True

    encoded_datasets = [pd.read_csv(p, index_col=0) for p in train_paths]

    X_train_list, X_test_list = \
        [df.loc[indcs_train, :].iloc[:, :-1].values
         for df in encoded_datasets], \
        [df.loc[indcs_test, :].iloc[:, :-1].values
         for df in encoded_datasets]

    y_train, y_test = \
        encoded_datasets[0].loc[indcs_train, "y"].values, \
        encoded_datasets[0].loc[indcs_test, "y"].values

    clf = MODEL[wildcards.model]
    eclf = META_MODEL[wildcards.meta_model]
    eclf.estimators = [(train_paths[i], clf) for i in range(len(train_paths))]

    try:
        eclf.fit(X_train_list, y_train)
        y_pred = eclf.predict(X_test_list)
        mcc = matthews_corrcoef(y_test,y_pred)
    except np.linalg.LinAlgError as e:
        print(e)
    except ValueError as e:
        print(e)

    pd.DataFrame({
        "mcc": [mcc],
        "fold": [wildcards.fold],
        "model": [wildcards.model],
        "meta_model": [wildcards.meta_model]
    }).to_csv(output[0])

    df_points.to_csv(output[1])
SnakeMake From line 448 of main/Snakefile 
511
512
513
514
515
516
run:
    df_points = pd.read_csv(input[0], index_col=0)

    idcs = default_rng().choice(df_points.index, size=15, replace=False)

    pd.DataFrame(idcs, columns=["enc_index"]).to_csv(output[0])
SnakeMake From line 511 of main/Snakefile 
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
run:
    df_indcs_train = pd.read_csv(input[0], index_col=0)
    indcs_train_tmp = df_indcs_train[f"fold_{wildcards.fold}"]
    df_indcs_val = pd.read_csv(input[1],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]
    indcs_train = pd.concat([indcs_train_tmp, indcs_val])

    df_indcs_test = pd.read_csv(input[2], index_col=0)
    indcs_test = df_indcs_test[f"fold_{wildcards.fold}"]

    df_points = pd.read_csv(input[3], index_col=0)

    idcs = pd.read_csv(input[4], index_col=0)["enc_index"]
    train_paths = list(set(
        df_points.iloc[idcs, :][["encoding_1", "encoding_2"]].values.flatten()
    ))

    # keep ensemble best encodings position for later usage
    df_points["ensemble_rand"] = False
    df_points.iloc[idcs, df_points.columns.get_loc("ensemble_rand")] = True

    encoded_datasets = [pd.read_csv(p, index_col=0) for p in train_paths]

    X_train_list, X_test_list = \
        [df.loc[indcs_train, :].iloc[:, :-1].values
         for df in encoded_datasets], \
        [df.loc[indcs_test, :].iloc[:, :-1].values
         for df in encoded_datasets]

    y_train, y_test = \
        encoded_datasets[0].loc[indcs_train, "y"].values, \
        encoded_datasets[0].loc[indcs_test, "y"].values

    clf = MODEL[wildcards.model]
    eclf = META_MODEL[wildcards.meta_model]
    eclf.estimators = [(train_paths[i], clf) for i in range(len(train_paths))]

    try:
        eclf.fit(X_train_list,y_train)
        y_pred = eclf.predict(X_test_list)
        mcc = matthews_corrcoef(y_test,y_pred)
    except np.linalg.LinAlgError as e:
        print(e)
    except ValueError as e:
        print(e)

    pd.DataFrame({
        "mcc": [mcc],
        "fold": [wildcards.fold],
        "model": [wildcards.model],
        "meta_model": [wildcards.meta_model]
    }).to_csv(output[0])

    df_points.to_csv(output[1])
SnakeMake From line 528 of main/Snakefile 
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
run:
    df_indcs_train = pd.read_csv(input[0], index_col=0)
    indcs_train_tmp = df_indcs_train[f"fold_{wildcards.fold}"]
    df_indcs_val = pd.read_csv(input[1],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]
    indcs_train = pd.concat([indcs_train_tmp, indcs_val])

    df_indcs_test = pd.read_csv(input[2], index_col=0)
    indcs_test = df_indcs_test[f"fold_{wildcards.fold}"]

    df_points = pd.read_csv(input[3], index_col=0)

    train_paths = list(set(
        df_points.loc[df_points.chull != -1][["encoding_1", "encoding_2"]]\
            .values.flatten()
    ))

    # keep ensemble best encodings position for later usage
    indices = df_points.loc[df_points.chull != -1].index
    df_points["ensemble_chull"] = False
    df_points.iloc[indices, df_points.columns.get_loc("ensemble_chull")] = True

    encoded_datasets = [pd.read_csv(p, index_col=0) for p in train_paths]

    X_train_list, X_test_list = \
        [df.loc[indcs_train, :].iloc[:, :-1].values
         for df in encoded_datasets], \
        [df.loc[indcs_test, :].iloc[:, :-1].values
         for df in encoded_datasets]

    y_train, y_test = \
        encoded_datasets[0].loc[indcs_train, "y"].values, \
        encoded_datasets[0].loc[indcs_test, "y"].values

    clf = MODEL[wildcards.model]
    eclf = META_MODEL[wildcards.meta_model]
    eclf.estimators = [(train_paths[i], clf) for i in range(len(train_paths))]

    try:
        eclf.fit(X_train_list,y_train)
        y_pred = eclf.predict(X_test_list)
        mcc = matthews_corrcoef(y_test,y_pred)
    except np.linalg.LinAlgError as e:
        print(e)
    except ValueError as e:
        print(e)

    pd.DataFrame({
        "mcc": [mcc],
        "fold": [wildcards.fold],
        "model": [wildcards.model],
        "meta_model": [wildcards.meta_model]
    }).to_csv(output[0])

    df_points.to_csv(output[1])
SnakeMake From line 592 of main/Snakefile 
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
run:
    df_indcs_train = pd.read_csv(input[0], index_col=0)
    indcs_train_tmp = df_indcs_train[f"fold_{wildcards.fold}"]
    df_indcs_val = pd.read_csv(input[1],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]
    indcs_train = pd.concat([indcs_train_tmp, indcs_val])

    df_indcs_test = pd.read_csv(input[2], index_col=0)
    indcs_test = df_indcs_test[f"fold_{wildcards.fold}"]

    df_points = pd.read_csv(input[3], index_col=0)

    train_paths = list(set(
        df_points.loc[df_points.pfront != -1][["encoding_1", "encoding_2"]] \
            .values.flatten()
    ))

    # keep ensemble best encodings position for later usage
    indices = df_points.loc[df_points.pfront != -1].index
    df_points["ensemble_pfront"] = False
    df_points.iloc[indices, df_points.columns.get_loc("ensemble_pfront")] = True

    encoded_datasets = [pd.read_csv(p, index_col=0) for p in train_paths]

    X_train_list, X_test_list = \
        [df.loc[indcs_train, :].iloc[:, :-1].values
         for df in encoded_datasets], \
        [df.loc[indcs_test, :].iloc[:, :-1].values
         for df in encoded_datasets]

    y_train, y_test = \
        encoded_datasets[0].loc[indcs_train, "y"].values, \
        encoded_datasets[0].loc[indcs_test, "y"].values

    clf = MODEL[wildcards.model]
    eclf = META_MODEL[wildcards.meta_model]
    eclf.estimators = [(train_paths[i], clf) for i in range(len(train_paths))]

    try:
        eclf.fit(X_train_list,y_train)
        y_pred = eclf.predict(X_test_list)
        mcc = matthews_corrcoef(y_test,y_pred)
    except np.linalg.LinAlgError as e:
        print(e)
    except ValueError as e:
        print(e)

    pd.DataFrame({
        "mcc": [mcc],
        "fold": [wildcards.fold],
        "model": [wildcards.model],
        "meta_model": [wildcards.meta_model]
    }).to_csv(output[0])

    df_points.to_csv(output[1])
SnakeMake From line 657 of main/Snakefile 
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
run:
    # use complete for MVO inner cv
    df_indcs_train = pd.read_csv(input[0], index_col=0)
    indcs_train_tmp = df_indcs_train[f"fold_{wildcards.fold}"]
    df_indcs_val = pd.read_csv(input[1],index_col=0)
    indcs_val = df_indcs_val[f"fold_{wildcards.fold}"]
    indcs_train = pd.concat([indcs_train_tmp, indcs_val])

    # use for testing after optimization
    df_indcs_test = pd.read_csv(input[2], index_col=0)
    indcs_test = df_indcs_test[f"fold_{wildcards.fold}"]

    df_points = pd.read_csv(input[3], index_col=0)

    # y is average pairwise error
    train_paths = list(set(
        df_points[["encoding_1", "encoding_2"]] \
            .values.flatten()
    ))

    n_universes = 32
    max_generations = 15

    p_0 = 6 / len(train_paths)
    mvo = BinaryMVO(
        n_universes=n_universes,
        d=len(train_paths),
        f=ff.train_ensemble,
        f_args={
            "paths_to_encoded_datasets": train_paths,
            "train_index": indcs_train,
            "base_clf": MODEL[wildcards.model],
            "meta_clf": META_MODEL[wildcards.meta_model]
        },
        p=[p_0, 1 - p_0],
        funker_name=None,
        new_random_state_each_generation=False,
        n_jobs=n_universes,
        log_path=os.path.dirname(output[2]) + "/",
        log_file_name=os.path.basename(output[2])
    )

    best_solution, _ = mvo.run(0, max_iterations=max_generations, parallel=True)

    train_paths_best = np.array(train_paths)[np.nonzero(best_solution)[0]]

    # keep ensemble best encodings position for later usage
    indices = df_points.loc[
        df_points.encoding_1.isin(train_paths_best) &
        df_points.encoding_2.isin(train_paths_best)
    ].index
    df_points["ensemble_mvo"] = False
    df_points.iloc[indices, df_points.columns.get_loc("ensemble_mvo")] = True

    encoded_datasets = [pd.read_csv(p, index_col=0) for p in train_paths_best]

    X_train_list, X_test_list = \
        [df.loc[indcs_train, :].iloc[:, :-1].values
         for df in encoded_datasets], \
        [df.loc[indcs_test, :].iloc[:, :-1].values
         for df in encoded_datasets]

    y_train, y_test = \
        encoded_datasets[0].loc[indcs_train, "y"].values, \
        encoded_datasets[0].loc[indcs_test, "y"].values

    clf = MODEL[wildcards.model]
    eclf = META_MODEL[wildcards.meta_model]
    eclf.estimators = [(train_paths[i], clf) for i in range(len(train_paths_best))]

    try:
        eclf.fit(X_train_list, y_train)
        y_pred = eclf.predict(X_test_list)
        mcc = matthews_corrcoef(y_test,y_pred)
    except np.linalg.LinAlgError as e:
        print(e)
    except ValueError as e:
        print(e)

    pd.DataFrame({
        "mcc": [mcc],
        "fold": [wildcards.fold],
        "model": [wildcards.model],
        "meta_model": [wildcards.meta_model]
    }).to_csv(output[0])

    df_points.to_csv(output[1])
SnakeMake From line 725 of main/Snakefile 
833
834
run:
    combine_point_data(list(input), output[0])
SnakeMake From line 833 of main/Snakefile 
844
845
run:
    combine_point_data(list(input), output[0])
SnakeMake From line 844 of main/Snakefile 
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
run:
    df_res = pd.DataFrame()
    for k, path_obj in input.items():
        if type(path_obj) == snakemake.io.Namedlist:
            for p in path_obj:
                df_tmp = pd.read_csv(p, index_col=0)
                df_tmp["cat"] = k
                df_res = pd.concat([df_res, df_tmp])
        else:
            df_tmp = pd.read_csv(path_obj, index_col=0)
            df_tmp["meta_model"] = wildcards.meta_model
            df_tmp["cat"] = k
            df_tmp.drop("encoding",axis=1,inplace=True)
            df_res = pd.concat([df_res, df_tmp])

    df_res.to_csv(output[0])
SnakeMake From line 871 of main/Snakefile 
895
896
897
898
899
900
901
run:
    df_res = pd.DataFrame()
    for  p in list(input):
        df_tmp = pd.read_csv(p, index_col=0)
        df_res = pd.concat([df_res, df_tmp])

    df_res.reset_index(drop=True).to_csv(output[0])
SnakeMake From line 895 of main/Snakefile 
911
912
script:
    "scripts/cd.R"
SnakeMake From line 911 of main/Snakefile 
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
run:
    # we use only one ensemble method here, because it does not influence the kappa-error values
    df_res = pd.DataFrame()
    for m in MODELS:
        for f in FOLDS:
            path = [p for p in list(input) if f"/{m}/" in p and f"/{f}.csv" in p][0]
            df_tmp = pd.read_csv(path, index_col=0)
            filter_vals = [
                df_tmp.ensemble_best,
                df_tmp.ensemble_rand,
                df_tmp.ensemble_chull,
                df_tmp.ensemble_pfront
            ]
            if "ensemble_mvo" in df_tmp.columns:
                filter_vals.append(df_tmp.ensemble_mvo)
            filter_ = reduce(
                lambda v1, v2: v1 | v2,
                filter_vals[:-1],
                filter_vals[-1]
            )
            df_tmp1 = df_tmp \
                .loc[np.bitwise_not(filter_) & (df_tmp.chull_complete == -1)] \
                .sample(1000).copy()
            df_tmp2 = df_tmp \
                .loc[filter_ | (df_tmp.chull_complete != -1)].copy()
            df_tmp = pd.concat([df_tmp1, df_tmp2])
            df_tmp["model"], df_tmp["fold"] = m, f
            df_res = pd.concat([df_res, df_tmp])

    df_res.loc[df_res.ensemble_mvo.isna(), "ensemble_mvo"] = False

    df_res["cat"] = df_res.apply(
        lambda row:
            "mvo" if row.ensemble_mvo else
            "chull" if row.ensemble_chull else
            "pfront" if row.ensemble_pfront else
            "best" if row.ensemble_best else
            "rand" if row.ensemble_rand else
            "all"
        , axis=1)

    df_res.to_csv(output[0])
SnakeMake From line 921 of main/Snakefile 
969
970
script:
    "scripts/plots/kappa_error.py"
SnakeMake From line 969 of main/Snakefile 
980
981
script:
    "scripts/plots/gens_vs_perf.py"
SnakeMake From line 980 of main/Snakefile 
994
995
script:
    "scripts/plots/box_plot.py"
SnakeMake From line 994 of main/Snakefile 
1006
1007
script:
    "scripts/plots/xcd.py"
SnakeMake From line 1006 of main/Snakefile 
1014
1015
script:
    "scripts/plots/box_plot_manova.py"
SnakeMake From line 1014 of main/Snakefile 
1031
1032
script:
    "scripts/statistics.R"
SnakeMake From line 1031 of main/Snakefile 
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
run:
    with open(output[0], "w") as f:
        for p in sorted(input):
            df_tmp = pd.read_csv(
                p,index_col=0,
                converters={"term": lambda v: v.replace("df_res$","")}
            )
            df_tmp.fillna("-",inplace=True)
            exp = p.split("/")[-1][:-4]
            df_tmp["experiment"] = exp
            f.write(f"<h4>{exp}</h4>\n{df_tmp.to_html(col_space='70px')}\n")
SnakeMake From line 1046 of main/Snakefile 
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
run:
    paths = [p.split("/")[-1].replace(".csv", "") for p in list(input)]
    paths = sorted(set(paths))
    arr = ["electrostatic_hull", "dist_freq"]
    paths = [[p[:18]] + p[18:].split("_")
             if "hull" in p else [p[:9]] + p[10:].split("_")
             if "dist_freq" in p else p.split("_") for p in paths]
    paths = [{k: v for k,v in zip(range(len(p)), p)} for p in paths]

    df = pd.DataFrame(paths)
    df.columns = ["param_" + str(i) for i in range(5)]


    def calc(vals):
        return "; ".join([str(v) for v in set(vals)])


    df = df.groupby("param_0").apply(lambda df: pd.DataFrame({
        "params_1": [calc(df["param_1"].values)],
        "params_2": [calc(df["param_2"].values)],
        "params_3": [calc(df["param_3"].values)],
        "params_4": [calc(df["param_4"].values)],
        #"params_5": [calc(df["param_5"].values)],
    })).reset_index(drop=False)

    df = df.drop(["level_1"], axis=1)

    cols = list(df.columns)
    cols[0] = "encoding"
    df.columns = cols

    df = df.replace("nan", "")

    df.to_csv(output[0], sep=",", index_label=False)
SnakeMake From line 1065 of main/Snakefile 
1111
1112
shell:
    "tar czf {output[0]} {input}"
SnakeMake From line 1111 of main/Snakefile 
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
run:
    from scipy.stats import ttest_rel

    def get_table(df_res):

        df_stats = df_res \
            .groupby(["dataset", "model", "cat", "meta_model"])["mcc"] \
            .describe().reset_index() \
            .loc[:, ['dataset', 'model', 'cat', 'meta_model', 'mean', 'std']]

        df_final = df_stats \
            .groupby(["dataset", "cat"]) \
            .apply(lambda df: df.sort_values("mean", ascending=False).iloc[0, :]) \
            .reset_index(drop=True)

        df_final["anno"] = df_final[["mean", "std"]].apply(
            lambda row: f"{np.round(row[0], 2)} (±{np.round(row[1], 2)})",
            axis=1
        )

        df_out = df_final.pivot(index="dataset", columns="cat", values="anno")

        for ds in df_final.dataset.unique():

            best_ens_cat, best_ens_mm, best_ens_m = df_final\
                .loc[df_final.dataset == ds]\
                .sort_values("mean", ascending=False)[["cat", "meta_model", "model"]]\
                .iloc[0]

            single_best_mm, single_best_m = df_final\
                .loc[(df_final.dataset == ds) & (df_final.cat == "single")]\
                .sort_values("mean",ascending=False)[["meta_model", "model"]] \
                .iloc[0]

            a1 = df_res.loc[
                (df_res.dataset == ds) &
                (df_res.cat == best_ens_cat) &
                (df_res.meta_model == best_ens_mm) &
                (df_res.model == best_ens_m)
            , "mcc"].values

            a2 = df_res.loc[
                (df_res.dataset == ds) &
                (df_res.cat == "single") &
                (df_res.meta_model == single_best_mm) &
                (df_res.model == single_best_m)
            , "mcc"
            ].values[:len(a1)]  # in case MVO is best method

            _, pval = ttest_rel(a1, a2, alternative="greater")

            # 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
            if pval == 0:
                sig = "***"
            elif pval < 0.001:
                sig = "**"
            elif pval < 0.01:
                sig = "*"
            elif pval < 0.05:
                sig = "."
            else:
                sig = "ns"

            df_out.loc[ds, best_ens_cat] = \
                df_out.loc[ds, best_ens_cat].replace(" (",f"<sup>{sig}</sup> (")

            df_out.columns.name = None
            df_out.index.name = None

        return df_out.to_html(escape=False)

    df_res = pd.DataFrame()

    for p in list(input.ensembles_res):
        df_tmp = pd.read_csv(p,index_col=0)
        dataset = p.split("/")[2]
        df_tmp["dataset"] = dataset
        df_res = pd.concat([df_res, df_tmp])

    for p in list(input.single_encodings_res):
        df_tmp = pd.read_csv(p,index_col=0)
        dataset = p.split("/")[2]
        df_tmp["dataset"] = dataset
        df_tmp = df_tmp.loc[df_tmp["rank"] == "Top_1", :]
        df_tmp = df_tmp.drop(["encoding"],axis=1)
        df_tmp = df_tmp.rename(columns={"rank": "meta_model"})
        df_res = pd.concat([df_res, df_tmp])

    t1 = get_table(df_res)

    df_res = df_res.loc[df_res.model != "rf"]
    t2 = get_table(df_res)

    with open(output[0], "w") as f:
        h1 = "<h3>With RF</h3>"
        h2 = "<h3>Without RF</h3>"
        f.write(f"{h1}\n{t1}\n{h2}\n{t2}\n")
        f.flush()
SnakeMake scipy From line 1124 of main/Snakefile 
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
run:
    df_stats = pd.DataFrame()
    for p in list(input):
        df_tmp = pd.read_csv(p, index_col=0)
        df_res = df_tmp.area.describe()[["mean", "std"]]
        df_res["dataset"] = p.split("/")[2]
        df_res["model"] = p.split("/")[-2]
        df_stats = pd.concat([
            df_stats,
            df_res.to_frame().transpose()
        ])

    df_stats.reset_index(drop=True, inplace=True)

    df_stats["anno"] = df_stats[["mean", "std"]].apply(
        lambda row: f"{np.round(row[0], 2)} (±{np.round(row[1], 3)})",
        axis=1
    )

    df_out = df_stats.pivot(index="dataset", columns="model", values="anno")
    df_out.columns.name = None
    df_out.index.name = None

    with open(output[0], "w") as f:
        f.write(f"{df_out.to_html()}\n")
        f.flush()
SnakeMake From line 1229 of main/Snakefile
ShowHide 34 more snippets with no or duplicated tags.

Login to post a comment if you would like to share your experience with this workflow.

Do you know this workflow well? If so, you can request seller status , and start supporting this workflow.

Free

Created: 1yr ago
Updated: 1yr ago
Maitainers: public
URL: https://github.com/spaenigs/ensemble-performance
Name: ensemble-performance
Version: 1
Badge:
workflow icon

Insert copied code into your website to add a link to this workflow.

Downloaded: 0
Copyright: Public Domain
License: MIT License
  • Future updates

Related Workflows

psychip_snakemake
Show Details View Workflow
ENCODE pipeline for histone marks developed for the psychENCODE project
public

public

psychip pipeline is an improved version of the ENCODE pipeline for histone marks developed for the psychENCODE project. The o...
raw sequence reads Alignment Sequence alignment report macs2 ucsc-bedclip bedGraphToBigWig BEDTools BWA Picard SAMtools Snakemake
Free
74
ncov_2
Show Details View Workflow
Near-real time tracking of SARS-CoV-2 in Connecticut
public

public

Repository containing scripts to perform near-real time tracking of SARS-CoV-2 in Connecticut using genomic data. This pipeli...
JSON nextclade Augur Biopython FOCUS Pandas Snakemake bs4 epiweeks geopy matplotlib numpy pycountry pycountry-convert uszipcode
Free
28
cellranger-snakemake-gke
Show Details View Workflow
snakemake workflow to run cellranger on a given bucket using gke.
public

public

A Snakemake workflow for running cellranger on a given bucket using Google Kubernetes Engine. The usage of this workflow ...
macs2 ucsc-bedclip bedGraphToBigWig BEDTools BWA Picard SAMtools Snakemake
Free
74
atlas
Show Details View Workflow
ATLAS - Three commands to start analyzing your metagenome data
public

public

Metagenome-atlas is a easy-to-use metagenomic pipeline based on snakemake. It handles all steps from QC, Assembly, Binning, t...
raw sequence reads Genome assembly Annotation track checkm2 gunc prodigal snakemake-wrapper-utils MEGAHIT Atlas BBMap Biopython BioRuby Bwa-mem2 cd-hit CheckM DAS Diamond eggNOG-mapper v2 MetaBAT 2 Minimap2 MMseqs MultiQC Pandas Picard pyfastx SAMtools SemiBin Snakemake SPAdes SqueezeMeta TADpole VAMB CONCOCT ete3 gtdbtk h5py networkx numpy plotly psutil utils metagenomics
Free
175
rna-seq-star-deseq2
Show Details View Workflow
RNA-seq workflow using STAR and DESeq2
public

public

This workflow performs a differential gene expression analysis with STAR and Deseq2. The usage of this workflow is described ...
raw sequence reads Gene expression profiling ID list snakemake-wrapper-utils biomaRt PCAtools BiocParallel BWA Cutadapt DESeq2 GFFutils MultiQC Pandas SAMtools Snakemake STAR cli stringr tidyverse RNA-Seq
Free
41
dna-seq-gatk-variant-calling
Show Details View Workflow
This Snakemake pipeline implements the GATK best-practices workflow
public

public

This Snakemake pipeline implements the GATK best-practices workflow for calling small germline variants. The usage of thi...
VCF raw sequence reads Variant calling genetic variants gatk rust-bio-tools snakemake-wrapper-utils tabix BCFtools BWA FastQC MultiQC Pandas Picard SAMtools Snakemake Trimmomatic Variant Effect Predictor (VEP) common matplotlib numpy seaborn DNA
Free
57