import streamlit as st
import torch
import numpy as np
from transformers import AutoTokenizer
from transformers import BertForSequenceClassification
st.set_page_config(layout='wide', initial_sidebar_state='expanded')
col1, col2= st.columns(2)
with col1:
st.title("FireWatch")
st.markdown("PREDICT WHETHER HEAT SIGNATURES AROUND THE GLOBE ARE LIKELY TO BE FIRES!")
st.markdown("Traing Code at:")
st.markdown("https://colab.research.google.com/drive/1-IfOMJ-X8MKzwm3UjbJbK6RmhT7tk_ye?usp=sharing")
st.markdown("Try the Model Yourself at:")
st.markdown("https://colab.research.google.com/drive/1GmweeQrkzs0OXQ_KNZsWd1PQVRLCWDKi?usp=sharing")
st.markdown("## Sample Table")
table_html = """
<table style="border-collapse: collapse; width: 100%;">
<tr style="border: 1px solid orange;">
<th style="border: 1px solid orange; font-weight: bold;">Category</th>
<th style="border: 1px solid orange; font-weight: bold;">Latitude, Longitude, Brightness, FRP</th>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Likely</td>
<td style="border: 1px solid orange;">-26.76123, 147.15512, 393.02, 203.63</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Likely</td>
<td style="border: 1px solid orange;">-26.7598, 147.14514, 361.54, 79.4</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Unlikely</td>
<td style="border: 1px solid orange;">-25.70059, 149.48932, 313.9, 5.15</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Unlikely</td>
<td style="border: 1px solid orange;">-24.4318, 151.83102, 307.98, 8.79</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Unlikely</td>
<td style="border: 1px solid orange;">-23.21878, 148.91298, 314.08, 7.4</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Likely</td>
<td style="border: 1px solid orange;">7.87518, 19.9241, 316.32, 39.63</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Unlikely</td>
<td style="border: 1px solid orange;">-20.10942, 148.14326, 314.39, 8.8</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Unlikely</td>
<td style="border: 1px solid orange;">7.87772, 19.9048, 304.14, 13.43</td>
</tr>
<tr style="border: 1px solid orange;">
<td style="border: 1px solid orange;">Likely</td>
<td style="border: 1px solid orange;">-20.79866, 124.46834, 366.74, 89.06</td>
</tr>
</table>
"""
st.markdown(table_html, unsafe_allow_html=True)
tree = """
<div class="pine-tree" style="width: 50%; margin: 0 auto;">
<div class="tree-top"></div>
<div class="tree-top2"></div>
<div class="tree-bottom">
<div class="trunk"></div>
</div>
</div>
<style>
.pine-tree {
width: 15vw;
height: 20vw;
position: relative;
display: flex;
justify-content: center;
align-items: center;
}
.tree-top {
width: 0;
height: 0;
border-left: 8vw solid transparent;
border-right: 8vw solid transparent;
border-bottom: 13vw solid green;
position: absolute;
top: 0;
left: 0;
right: 0;
margin: auto;
}
.tree-top2 {
width: 0;
height: 0;
border-left: 8vw solid transparent;
border-right: 8vw solid transparent;
border-bottom: 13vw solid green;
position: absolute;
top: 3vw;
left: 0;
right: 0;
margin: auto;
}
.tree-bottom {
width: 8vw;
height: 10vw;
background-color: brown;
position: absolute;
bottom: 0;
left: 0;
right: 0;
top: 21vw;
margin: auto;
}
.trunk {
width: 3vw;
height: 10vw;
background-color: brown;
position: absolute;
bottom: 0;
left: 0;
right: 0;
margin: auto;
}
</style>
"""
with col2:
@st.cache(suppress_st_warning=True, allow_output_mutation=True)
def load_model(show_spinner=True):
MODEL_PATH = "NimaKL/FireWatch_tiny_75k"
model = BertForSequenceClassification.from_pretrained(MODEL_PATH)
return model
token_id = []
attention_masks = []
def preprocessing(input_text, tokenizer):
'''
Returns <class transformers.tokenization_utils_base.BatchEncoding> with the following fields:
- input_ids: list of token ids
- token_type_ids: list of token type ids
- attention_mask: list of indices (0,1) specifying which tokens should considered by the model (return_attention_mask = True).
'''
return tokenizer.encode_plus(
input_text,
add_special_tokens = True,
max_length = 16,
pad_to_max_length = True,
return_attention_mask = True,
return_tensors = 'pt'
)
def predict(new_sentence):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# We need Token IDs and Attention Mask for inference on the new sentence
test_ids = []
test_attention_mask = []
# Apply the tokenizer
encoding = preprocessing(new_sentence, tokenizer)
# Extract IDs and Attention Mask
test_ids.append(encoding['input_ids'])
test_attention_mask.append(encoding['attention_mask'])
test_ids = torch.cat(test_ids, dim = 0)
test_attention_mask = torch.cat(test_attention_mask, dim = 0)
# Forward pass, calculate logit predictions
with torch.no_grad():
output = model(test_ids.to(device), token_type_ids = None, attention_mask = test_attention_mask.to(device))
prediction = 'Likely' if np.argmax(output.logits.cpu().numpy()).flatten().item() == 1 else 'Unlikely'
pred = 'Predicted Class: '+ prediction
return pred
model = load_model()
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
with col2:
st.markdown('## Enter Prediction Data in Correct Format "Latitude, Longtitude, Brightness, FRP"')
text = st.text_input('Predition Data: ', 'Example: 8.81064, -65.07661, 328.04, 18.76')
aButton = st.button('Predict')
if text or aButton:
with st.spinner('Wait for it...'):
st.success(predict(text))
st.markdown(tree, unsafe_allow_html=True)