jiro_nn 0.8.1

Neural Networks framework with model building & data preprocessing features.
Documentation 
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

use std::collections::{HashMap, HashSet};

use crate::{
    dataset::{Dataset, Feature},
    datatable::DataTable,
    linalg::Scalar,
};

use super::{CachedConfig, DataTransformation};

pub struct OneHotEncode;

impl DataTransformation for OneHotEncode {
    fn transform(
        &mut self,
        _cached_config: &CachedConfig,
        dataset_config: &Dataset,
        data: &DataTable,
    ) -> (Dataset, DataTable) {
        let mut features_values: HashMap<Feature, HashSet<i64>> = HashMap::new();

        data.as_scalar_hashmap().iter().for_each(|(name, values)| {
            let feature = dataset_config.features.iter().find(|f| f.name == *name).unwrap();
            if feature.one_hot_encoded {
                let mut values_set = HashSet::new();
                values.iter().for_each(|v| {
                    values_set.insert(*v as i64);
                });
                features_values.insert(feature.clone(), values_set);
            }
        });

        let mut new_config = dataset_config.clone();

        for (feature, values) in features_values.iter() {
            for value in values.iter() {
                let mut new_feature = feature.clone();
                new_feature.name = format!("{}={}", feature.name, value);
                new_feature.one_hot_encoded = false;
                new_config = new_config.with_added_feature(new_feature);
            }
            new_config = new_config.without_feature(feature.name.clone());
        }

        let mut new_data = data.clone();
        for (feature, values) in features_values.iter() {
            let column = new_data.column_to_vector(&feature.name);
            new_data = new_data.drop_column(&feature.name);
            let mut rows = vec![vec![0.0 as Scalar; values.len()]; column.len()];
            let mut names = vec![];

            for (i, value) in values.iter().enumerate() {
                for (row, v) in column.iter().enumerate() {
                    if *v as i64 == *value {
                        rows[row][i] = 1.0;
                    } else {
                        rows[row][i] = 0.0;
                    }
                }
                names.push(format!("{}={}", feature.name, value));
            }

            let onehotdata = DataTable::from_vectors(names.as_slice(), &rows);
            new_data = new_data.append_table_as_columns(&onehotdata);
        }

        (new_config, new_data)
    }

    fn reverse_columnswise(&mut self, data: &DataTable) -> DataTable {
        let mut classes: HashMap<String, Vec<String>> = HashMap::new();
        let mut one_hot_encoded: HashMap<String, Vec<(usize, Scalar)>> = HashMap::new();

        // extract any column name like xxxx=yyyy which is a one hot encoded value for yyyy of column xxxx
        data.as_scalar_hashmap().iter().for_each(|(name, values)| {
            let parts: Vec<&str> = name.split("=").collect();
            if parts.len() == 2 {
                let column_name = parts[0].to_string();
                let class = parts[1].to_string();

                if !one_hot_encoded.contains_key(&column_name) {
                    classes.insert(column_name.clone(), vec![class.clone()]);
                    let class_idx = 0;

                    let mut idmaxes = Vec::new();
                    for i in 0..values.len() {
                        idmaxes.push((class_idx, values[i]));
                    }
                    one_hot_encoded.insert(column_name, idmaxes);
                } else {
                    let column_classes = classes.get_mut(&column_name).unwrap();
                    let class_idx = column_classes.len();
                    column_classes.push(class.clone());

                    let idmaxes = one_hot_encoded.get_mut(&column_name).unwrap();
                    for i in 0..values.len() {
                        idmaxes[i].1 = idmaxes[i].1.max(values[i]);
                        if idmaxes[i].1 == values[i] {
                            idmaxes[i].0 = class_idx;
                        }
                    }
                }
            }
        });

        let mut new_data = data.clone();

        // remove old column=class columns for all classes in classes for that column
        for (column_name, classes) in classes.iter() {
            for class in classes.iter() {
                new_data = new_data.drop_column(&format!("{}={}", column_name, class));
            }
        }

        // add new column columns for all columns and put the most likely class in there
        for (column_name, idmaxes) in one_hot_encoded.iter() {
            let mut column_class_values = Vec::new();
            let mut column_confidence_value = Vec::new();
            for (class_idx, confidence) in idmaxes {
                let class_name = classes.get(column_name).unwrap()[*class_idx].clone();
                column_class_values.push(class_name);
                column_confidence_value.push(*confidence);
            }
            new_data = new_data.with_column_string(&column_name, column_class_values.as_slice());
            new_data = new_data.with_column_scalar(&format!("{}.confidence", column_name), column_confidence_value.as_slice());
        }
        
        new_data
    }

    fn get_name(&self) -> String {
        "onehotencode".to_string()
    }
}