prrao87

hyper_params = { 
    "lr": 0.35,         # Learning rate
    "epoch": 100,       # Number of training epochs to train for
    "wordNgrams": 3,    # Number of word n-grams to consider during training
    "dim": 155,         # Size of word vectors
    "ws": 5,            # Size of the context window for CBOW or skip-gram
    "minn": 2,          # Min length of char ngram
    "maxn": 5,          # Max length of char ngram
    "bucket": 2014846,  # Number of buckets
}

prrao87

# Quantize model to reduce space usage
model.quantize(input=train, qnorm=True, retrain=True, cutoff=100000)
model.save_model(os.path.join(model_path, "sst-5.ftz"))

prrao87

class FastTextSentiment(Base):
    """Predict fine-grained sentiment scores using FastText"""
    def __init__(self, model_file: str=None) -> None:
        super().__init__()
        import fasttext
        self.model = fasttext.load_model(model_file)

    def score(self, text: str) -> int:
        # Predict just the top label (hence 1 index below)
        labels, probabilities = self.model.predict(text, 1)

prrao87

from flair.embeddings import FlairEmbeddings, BertEmbeddings, WordEmbeddings, DocumentRNNEmbeddings

# init Flair embeddings
flair_forward_embedding = FlairEmbeddings('multi-forward')
flair_backward_embedding = FlairEmbeddings('multi-backward')

# init BERT base (cases)
optional_embedding = BertEmbeddings('bert-base-cased')
# OR init ELMo (original)
# optional_embedding = ELMoEmbeddings('original')

prrao87

class FlairSentiment(Base):
    """Predict fine-grained sentiment scores using Flair."""
    def __init__(self, model_file: str=None) -> None:
        super().__init__()
        from flair.models import TextClassifier
        self.model = TextClassifier.load(model_file)

    def score(self, text: str) -> int:
        from flair.data import Sentence
        doc = Sentence(text)

prrao87

class ExampleExplainer:
    """Class to explain classification results.
       The `predict` method outputs a numpy array of floats, which
       is the classifier's prediction probability for each class.
    """
    def __init__(self, path_to_model: str) -> None:
        # Load in a trained classifier model
        
    def predict(self, texts: List[str]) -> np.array([float, ...]):
        # Take in a list of strings (LIME text sample of variations)

prrao87

def explainer(method: str, path_to_file: str, text: str, num_samples: int) -> LimeTextExplainer:
    """Run LIME explainer on provided classifier"""
    model = explainer_class(method, path_to_file)
    predictor = model.predict

    # Create a LimeTextExplainer
    explainer = LimeTextExplainer(
        # Specify split option for string
        split_expression=lambda x: x.split(),
        # Our classifer uses bigrms, trigrams or contextual ordering to classify text

prrao87

# First, offset the float score from the range [-1, 1] to a range [0, 1]
offset = (score(text) + 1) / 2.
# Convert offset float score in [0, 1] to an integer value in the range [1, 5]
binned = np.digitize(5 * offset, np.array([1, 2, 3, 4, 5])) + 1
# Similate probabilities of each class based on a normal distribution
simulated_probs = scipy.stats.norm.pdf(np.array([1, 2, 3, 4, 5]), binned, scale=0.5)

prrao87

#!/usr/bin/bash
fasttext supervised -input data/sst/sst_train.txt -output model_hyperopt \
-autotune-validation data/sst/sst_dev.txt -autotune-modelsize 10M -verbose 3

prrao87

import pandas as pd

TEXT_COL, LABEL_COL = 'text', 'truth'

def read_sst5(data_dir, colnames=[LABEL_COL, TEXT_COL]):
    datasets = {}
    for t in ["train", "dev", "test"]:
        df = pd.read_csv(os.path.join(data_dir, f"sst_{t}.txt"), sep='\t', header=None, names=colnames)
        df[LABEL_COL] = df[LABEL_COL].str.replace('__label__', '')
        df[LABEL_COL] = df[LABEL_COL].astype(int)   # Categorical data type for truth labels