DataCamp_Cosmetic.py

# Create a function to vetorize all the ingredients and get t-SNE at once
def cosmetic_map(option_1, option_2):
    ''' Define a function creating a dataframe for each option '''
    df = cosm[cosm['Label'] == option_1][cosm[option_2] == 1]
    df = df.reset_index()

    # embedding each ingredients
    word_index_map = {}
    index_word_map = []
    current_index = 0
    corpus = []

    for i in range(len(df)):
        text = df['Ingredients'][i]
        text = text.lower()
        tokens = text.split(', ')
        corpus.append(tokens)
        for token in tokens:
            if token not in word_index_map:
                word_index_map[token] = current_index
                current_index += 1
                index_word_map.append(token)

    # creating dtm matrix
    D = len(corpus)   # number of items
    N = len(word_index_map)   # total number of ingredients
    A = np.zeros((D, N))

    def tokens_to_vector(tokens):
        x = np.zeros(len(word_index_map))
        for token in tokens:
            i = word_index_map[token]
            x[i] = 1
        return x

    i = 0
    for tokens in corpus:
        A[i, :] = tokens_to_vector(tokens)
        i += 1

    # decomposition using t-SNE
    tsne = TSNE(learning_rate = 200)
    tsne_features = tsne.fit_transform(A)
    df['X'] = tsne_features[:, 0]
    df['Y'] = tsne_features[:, 1]

    return df


# Apply the function to all combination of the two options. 
df_all = pd.DataFrame()
for a in option_1:
    for b in option_2:
            temp = cosmetic_map(a, b)
            temp['Label'] = a + '_' + b
            df_all = pd.concat([df, temp], axis = 0)