File size: 7,812 Bytes
99d1428
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
import streamlit as st
import torch
from transformers import AlbertTokenizer, AlbertForSequenceClassification, AlbertModel
import numpy as np
import pandas as pd
import os
from torch.nn.functional import softmax
import torch.nn as nn

# Paths
LEVEL_DIRS = {
    1: 'level1',
    2: 'level2',
    3: 'level3',
    4: 'level4',
    5: 'level5',
    6: 'level6',
    7: 'level7'
}
MAPPING_FILE = 'mapping.csv'
MODEL_NAME = 'albert/albert-base-v2'  # Define the base model name

# Load mapping
mapping_df = pd.read_csv(MAPPING_FILE)

def get_label_text(level, predicted_id):
    level_map = {0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6}
    level_num = level_map.get(level)
    if level_num is not None:
        row = mapping_df[(mapping_df['level'] == level_num) & (mapping_df['id'] == predicted_id)]
        return row['text'].iloc[0] if not row.empty else "Description not found"
    return "Invalid Level"

def predict_level(level, text, parent_prediction_id=None, checkpoint_path=None):
    level_dir = LEVEL_DIRS[level]
    tokenizer = AlbertTokenizer.from_pretrained(checkpoint_path)
    label_map = np.load(os.path.join(level_dir, 'label_map.npy'), allow_pickle=True).item()
    num_labels = len(label_map)

    if level == 1:
        model = AlbertForSequenceClassification.from_pretrained(checkpoint_path)
    else:
        parent_level_dir = LEVEL_DIRS[level - 1]
        parent_label_map = np.load(os.path.join(parent_level_dir, 'label_map.npy'), allow_pickle=True).item()
        num_parent_labels = len(parent_label_map)

        class TaxonomyClassifier(nn.Module):
            def __init__(self, base_model_name, num_parent_labels, num_labels):
                super().__init__()
                self.albert = AlbertModel.from_pretrained(base_model_name)
                self.dropout = nn.Dropout(0.1)
                self.classifier = nn.Linear(self.albert.config.hidden_size + num_parent_labels, num_labels)

            def forward(self, input_ids, attention_mask, parent_ids):
                outputs = self.albert(input_ids, attention_mask=attention_mask)
                pooled_output = outputs.pooler_output
                pooled_output = self.dropout(pooled_output)
                combined_features = torch.cat((pooled_output, parent_ids), dim=1)
                logits = self.classifier(combined_features)
                return logits

        model = TaxonomyClassifier(MODEL_NAME, num_parent_labels, num_labels)
        model.load_state_dict(torch.load(os.path.join(checkpoint_path, 'model.safetensors'), map_location=torch.device('cpu')))

    model.eval()
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)

    if level > 1:
        parent_label_map_current = np.load(os.path.join(LEVEL_DIRS[level - 1], 'label_map.npy'), allow_pickle=True).item()
        num_parent_labels_current = len(parent_label_map_current)
        parent_one_hot = torch.zeros(num_parent_labels_current)
        if parent_prediction_id != 0:
            parent_index = parent_label_map_current.get(parent_prediction_id)
            if parent_index is not None:
                parent_one_hot[parent_index] = 1.0
        with torch.no_grad():
            outputs = model(inputs.input_ids, attention_mask=inputs.attention_mask, parent_ids=parent_one_hot.unsqueeze(0))
    else:
        with torch.no_grad():
            outputs = model(**inputs)

    probabilities = softmax(outputs.logits if level == 1 else outputs, dim=-1)[0]
    top3_prob, top3_indices = torch.topk(probabilities, 3)
    index_to_label = {v: k for k, v in label_map.items()}
    results = []
    for prob, index in zip(top3_prob, top3_indices):
        predicted_label_id = index_to_label[index.item()]
        results.append((predicted_label_id, prob.item()))
    return results

st.title("Taxonomy Model Inference")

input_text = st.text_area("Enter text to classify", "Experience the magic of music with the Clavinova CLP-800 series. This versatile range of digital pianos is designed to delight everyone, from budding musicians to seasoned pianists. Each model combines state-of-the-art technology with the realistic touch and tone of world-renowned grand pianos, enhanced by GrandTouch keyboard action and Virtual Resonance Modeling. With seamless Bluetooth® connectivity, built-in lessons, and elegant design, the CLP-800 series offers the perfect blend of tradition and innovation. Elevate your musical journey with the warmth and sophistication of the Yamaha Clavinova, our finest series of digital pianos.")

softmax_threshold = st.slider("Softmax Threshold", min_value=0.0, max_value=1.0, value=0.5, step=0.05)

# Checkpoint Selection
available_levels = []
level_checkpoints = {}
for level in LEVEL_DIRS:
    level_dir = LEVEL_DIRS[level]
    if os.path.exists(level_dir):
        options = [d for d in os.listdir(level_dir) if os.path.isdir(os.path.join(level_dir, d))]
        options = [d for d in options if 'step' in d or d == 'model']
        options.sort(key=lambda x: (('step' not in x), int(x.split('step')[-1]) if 'step' in x else -1))
        level_checkpoints[level] = [os.path.join(level_dir, opt) for opt in options]
        if level_checkpoints[level]:
            available_levels.append(level)
    else:
        level_checkpoints[level] = []

selected_checkpoints = {}
for level in available_levels:
    selected_checkpoints[level] = st.selectbox(f"Select Level {level} Checkpoint", options=level_checkpoints[level])

if st.button("Run Inference"):
    if input_text:
        all_level_results = {}
        current_prediction_id = None
        last_level = 0

        for level in sorted(available_levels):
            if selected_checkpoints[level]:
                checkpoint_path = selected_checkpoints[level]
                if level == 1:
                    level_results = predict_level(level, input_text, checkpoint_path=checkpoint_path)
                else:
                    if current_prediction_id == 0:
                        st.info(f"Taxonomy terminated at Level {last_level} with ID 0.")
                        break
                    level_results = predict_level(level, input_text, parent_prediction_id=current_prediction_id, checkpoint_path=checkpoint_path)

                if level_results[0][1] < softmax_threshold:
                    st.info(f"Inference stopped at Level {level} due to softmax probability ({level_results[0][1]:.3f}) being below the threshold.")
                    break

                all_level_results[level] = level_results
                current_prediction_id = level_results[0][0]
                last_level = level
            else:
                st.warning(f"Skipping Level {level} as no checkpoint is selected.")
                break

        data = []
        for level in sorted(all_level_results.keys()):
            results = all_level_results[level]
            data.append({
                'level': level,
                'text': get_label_text(level - 1, results[0][0]),
                'softmax': f"{results[0][1]:.3f}",
                'runner_up_1_id': results[1][0],
                'runner_up_1_text': get_label_text(level - 1, results[1][0]),
                'runner_up_1_softmax': f"{results[1][1]:.3f}",
                'runner_up_2_id': results[2][0],
                'runner_up_2_text': get_label_text(level - 1, results[2][0]),
                'runner_up_2_softmax': f"{results[2][1]:.3f}",
            })

        if data:
            df = pd.DataFrame(data)
            st.dataframe(df)
        else:
            st.info("No predictions made or inference stopped.")

    else:
        st.warning("Please enter text for classification.")