doubt.datasets.fish_bioconcentration

  1"""Fish bioconcentration data set.
  2
  3This data set is from the UCI data set archive, with the description being the original
  4description verbatim. Some feature names may have been altered, based on the
  5description.
  6"""
  7
  8import io
  9import zipfile
 10
 11import pandas as pd
 12
 13from .dataset import BASE_DATASET_DESCRIPTION, BaseDataset
 14
 15
 16class FishBioconcentration(BaseDataset):
 17    __doc__ = f"""
 18    This dataset contains manually-curated experimental bioconcentration factor (BCF)
 19    for 1058 molecules (continuous values). Each row contains a molecule, identified by
 20    a CAS number, a name (if available), and a SMILES string. Additionally, the KOW
 21    (experimental or predicted) is reported. In this database, you will also find
 22    Extended Connectivity Fingerprints (binary vectors of 1024 bits), to be used as
 23    independent variables to predict the BCF.
 24
 25    {BASE_DATASET_DESCRIPTION}
 26
 27    Features:
 28        logkow (float):
 29            Octanol water paritioning coefficient (experimental or predicted, as
 30            indicated by ``KOW type``
 31        kow_exp (int):
 32            Indicates whether ``logKOW`` is experimental or predicted, with 1 denoting
 33            experimental and 0 denoting predicted
 34        smiles_[idx] for idx = 0..125 (int):
 35            Encoding of SMILES string to identify the 2D molecular structure. The
 36            encoding is as follows, where 'x' is a padding string to ensure that all
 37            the SMILES strings are of the same length:
 38
 39                - 0  = 'x'
 40                - 1  = '#'
 41                - 2  = '('
 42                - 3  = ')'
 43                - 4  = '+'
 44                - 5  = '-'
 45                - 6  = '/'
 46                - 7  = '1'
 47                - 8  = '2'
 48                - 9  = '3'
 49                - 10 = '4'
 50                - 11 = '5'
 51                - 12 = '6'
 52                - 13 = '7'
 53                - 14 = '8'
 54                - 15 = '='
 55                - 16 = '@'
 56                - 17 = 'B'
 57                - 18 = 'C'
 58                - 19 = 'F'
 59                - 20 = 'H'
 60                - 21 = 'I'
 61                - 22 = 'N'
 62                - 23 = 'O'
 63                - 24 = 'P'
 64                - 25 = 'S'
 65                - 26 = '['
 66                - 27 = '\\'
 67                - 28 = ']'
 68                - 29 = 'c'
 69                - 30 = 'i'
 70                - 31 = 'l'
 71                - 32 = 'n'
 72                - 33 = 'o'
 73                - 34 = 'r'
 74                - 35 = 's'
 75
 76    Targets:
 77        logbcf (float):
 78            Experimental fish bioconcentration factor (logarithm form)
 79
 80    Source:
 81        https://archive.ics.uci.edu/ml/datasets/QSAR+fish+bioconcentration+factor+%28BCF%29
 82
 83    Examples:
 84        Load in the data set::
 85
 86            >>> dataset = FishBioconcentration()
 87            >>> dataset.shape
 88            (1054, 129)
 89
 90        Split the data set into features and targets, as NumPy arrays::
 91
 92            >>> X, y = dataset.split()
 93            >>> X.shape, y.shape
 94            ((1054, 128), (1054,))
 95
 96        Perform a train/test split, also outputting NumPy arrays::
 97
 98            >>> train_test_split = dataset.split(test_size=0.2, random_seed=42)
 99            >>> X_train, X_test, y_train, y_test = train_test_split
100            >>> X_train.shape, y_train.shape, X_test.shape, y_test.shape
101            ((825, 128), (825,), (229, 128), (229,))
102
103        Output the underlying Pandas DataFrame::
104
105            >>> df = dataset.to_pandas()
106            >>> type(df)
107            <class 'pandas.core.frame.DataFrame'>
108    """
109
110    _url = (
111        "https://archive.ics.uci.edu/ml/machine-learning-databases/"
112        "00511/QSAR_fish_BCF.zip"
113    )
114
115    _features = range(128)
116    _targets = [128]
117
118    def _prep_data(self, data: bytes) -> pd.DataFrame:
119        """Prepare the data set.
120
121        Args:
122            data (bytes): The raw data
123
124        Returns:
125            Pandas dataframe: The prepared data
126        """
127        # Convert the bytes into a file-like object
128        buffer = io.BytesIO(data)
129
130        # Unzip the file and pull out the csv file
131        with zipfile.ZipFile(buffer, "r") as zip_file:
132            csv = zip_file.read("QSAR_BCF_Kow.csv")
133
134        # Convert the string into a file-like object
135        csv_file = io.BytesIO(csv)
136
137        # Read the file-like object into a dataframe
138        cols = ["cas", "name", "smiles", "logkow", "kow_exp", "logbcf"]
139        df = pd.read_csv(
140            csv_file,
141            names=cols,
142            header=0,
143            usecols=[col for col in cols if col not in ["cas", "name"]],
144        )
145
146        # Drop NaNs
147        df = df.dropna().reset_index(drop=True)
148
149        # Encode KOW types
150        kow_types = ["pred", "exp"]
151        df["kow_exp"] = df.kow_exp.map(lambda txt: kow_types.index(txt))
152
153        # Get maximum SMILE string length
154        max_smile = max(len(smile_string) for smile_string in df.smiles)
155
156        # Pad SMILE strings
157        df["smiles"] = [
158            smile_string + "x" * (max_smile - len(smile_string))
159            for smile_string in df.smiles
160        ]
161
162        # Split up the SMILE strings into a matrix
163        smile_df = pd.DataFrame(df.smiles.map(list).values.tolist())
164
165        # Set the column values of the SMILE dataframe
166        smile_df.columns = pd.Index(
167            [f"smiles_{idx}" for idx in range(smile_df.shape[1])]
168        )
169
170        # Add the smile dataframe to the original dataframe
171        df = pd.concat([df, smile_df], axis=1)
172
173        # Drop original SMILE feature
174        df = df.drop(columns="smiles")
175
176        # Put the target variable at the end
177        cols = ["logkow", "kow_exp"]
178        cols += [f"smiles_{idx}" for idx in range(max_smile)]
179        cols += ["logbcf"]
180        df = df[cols]
181
182        # Ensure that the `logkow` column is numeric
183        df["logkow"] = pd.to_numeric(df.logkow)
184
185        return df