Titanic classification by EXGBoost
Mix.install([
{:exgboost, "~> 0.3"},
{:scholar, "~> 0.2"},
{:csv, "~> 3.0"},
{:exla, "~> 0.6"},
{:statistics, "~> 0.6"},
{:kino_vega_lite, "~> 0.1"}
])
:ok
Section
defmodule Pre do
def count_missings(datas) do
datas
|> Enum.flat_map(fn map -> Map.filter(map, &(elem(&1, 1) == "")) |> Map.keys() end)
|> Enum.reject(&(&1 == []))
|> Enum.frequencies()
end
def separate(datas, id, label) do
{
datas |> Enum.map(&Map.get(&1, id)),
if Map.has_key?(List.first(datas), label) do
datas |> Enum.map(&([[String.to_float("#{Map.get(&1, label)}.0")]] |> Nx.tensor()))
else
nil
end,
datas |> Enum.map(&Map.drop(&1, [id, label]))
}
end
def drop(datas, keys) do
datas
|> Enum.map(&Map.drop(&1, keys))
end
def empty_replace(datas, replaces_map) do
replaces_map
|> Enum.reduce(datas, fn {key, replace}, acc ->
acc
|> Enum.map(&Map.put(&1, key, String.replace(Map.get(&1, key), ~r/^$/, replace)))
end)
end
def make_dummies(datas, key) do
datas
|> Enum.map(&Map.get(&1, key))
|> Enum.uniq()
|> Enum.with_index(&{&1, String.to_float("#{&2}.0")})
|> Enum.into(%{})
end
def to_dummies(datas, train_maps, keys) do
keys
|> Enum.reduce(datas, fn key, acc ->
acc
|> Enum.map(
&Map.put(
&1,
key,
case make_dummies(train_maps, key)[Map.get(&1, key)] do
nil -> 10.0
n -> n
end
)
)
end)
end
def integer_string_to_float(datas, keys) do
keys
|> Enum.reduce(datas, fn key, acc ->
acc
|> Enum.map(
&Map.put(
&1,
key,
Map.get(&1, key)
|> String.replace(~r/^(?!.*\.).*$/, "\\0\.0")
|> String.to_float()
)
)
end)
end
def map_to_tensor(datas) do
datas
|> Enum.map(&([Map.values(&1)] |> Nx.tensor()))
end
def for_dummies(datas) do
datas
|> Enum.map(
&Map.put(
&1,
:honor,
&1.name |> String.replace(~r/^.*, /, "") |> String.replace(~r/. .*/, "")
)
)
|> drop([:cabin, :name, :ticket])
|> empty_replace(%{embarked: "S", age: "30", fare: "32"})
end
def process(datas, train_datas) do
for_dummies_train_data = for_dummies(train_datas)
datas
|> for_dummies
|> to_dummies(for_dummies_train_data, [:embarked, :sex, :honor])
|> integer_string_to_float([:age, :fare, :parch, :pclass, :sibsp])
|> map_to_tensor
end
def header_and_csv_datas(path) do
[hd | tl] =
File.stream!(path)
|> CSV.decode!()
|> Enum.to_list()
header =
hd
|> Enum.map(&(&1 |> String.downcase() |> String.to_atom()))
{header, tl}
end
def csv_file_to_datas(path) do
{header, datas} = header_and_csv_datas(path)
datas
|> Enum.map(&(List.zip([header, &1]) |> Enum.into(%{})))
|> Pre.separate(:passengerid, :survived)
end
end
{train_csv_ids, train_csv_labels, train_csv_trains} = Pre.csv_file_to_datas("train.csv")
count = train_csv_trains |> Enum.count()
{trains, validate_trains} =
Pre.process(train_csv_trains, train_csv_trains) |> Enum.split(div(count, 10) * 8)
{labels, validate_labels} = train_csv_labels |> Enum.split(div(count, 10) * 8)
{[
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
...
]
>,
...
],
[
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[1.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
[0.0]
]
>,
#Nx.Tensor<
f32[1][1]
[
...
]
>,
...
]}
alias Scholar.Metrics.Classification
model = EXGBoost.train(trains, labels)
predicts =
EXGBoost.predict(model, validate_trains)
|> Nx.map([type: :f32], &Nx.greater(&1, 0.5))
y_validate = validate_labels |> Enum.map(&(&1 |> Nx.to_list())) |> List.flatten() |> Nx.tensor()
Classification.accuracy(predicts, y_validate)
#Nx.Tensor<
f32
0.8603351712226868
>
{test_csv_ids, _, test_csv_trains} = Pre.csv_file_to_datas("test.csv")
result =
Pre.process(test_csv_trains, train_csv_trains)
|> Enum.map(&(EXGBoost.predict(model, &1) |> Nx.to_flat_list() |> List.first() |> round))
|> then(&Enum.zip(test_csv_ids, &1))
|> Enum.map(&[elem(&1, 0), Integer.to_string(elem(&1, 1))])
|> then(&[["PassengerId", "Survived"] | &1])
[
["PassengerId", "Survived"],
["892", "0"],
["893", "0"],
["894", "0"],
["895", "0"],
["896", "1"],
["897", "0"],
["898", "0"],
["899", "0"],
["900", "1"],
["901", "0"],
["902", "0"],
["903", "0"],
["904", "1"],
["905", "0"],
["906", "1"],
["907", "1"],
["908", "0"],
["909", "0"],
["910", "1"],
["911", "0"],
["912", "0"],
["913", "0"],
["914", "1"],
["915", "1"],
["916", "1"],
["917", "0"],
["918", "1"],
["919", "0"],
["920", "0"],
["921", "0"],
["922", "0"],
["923", "0"],
["924", "1"],
["925", "0"],
["926", "0"],
["927", "0"],
["928", "0"],
["929", "0"],
["930", "0"],
["931", "1"],
["932", "0"],
["933", "1"],
["934", "0"],
["935", "1"],
["936", "1"],
["937", "0"],
["938", "0"],
["939", ...],
[...],
...
]
result
|> CSV.encode()
|> Enum.to_list()
|> then(&File.write("result.csv", &1))
:ok
