hi

app.py

@@ -1,25 +1,51 @@
 import json
+import pandas as pd
 import gradio as gr
+from transformers import PreTrainedTokenizerFast, BertForMaskedLM
+from datasets import load_dataset
+import xgboost_infer
 
+embeddings_train = load_dataset("LofiAmazon/BOLD-Embeddings-Ecolayers-Amazon", split='train').to_pandas()
 
 with open("default_inputs.json", "r") as default_inputs_file:
     DEFAULT_INPUTS = json.load(default_inputs_file)
 
-
 def set_default_inputs():
     return (DEFAULT_INPUTS["dna_sequence"],
             DEFAULT_INPUTS["latitude"],
             DEFAULT_INPUTS["longitude"])
 
+def preprocess(): 
+    ''' prepares app input for the genus prediction model
+    '''
+    # preprocess DNA seq
+    # Replace all symbols in nucraw which are not A, C, G, T with N
+    inp_dna = inp_dna.str.replace("[^ACGT]", "N", regex=True)
+    # Truncate trailing Ns from nucraw
+    inp_dna = inp_dna.str.replace("N+$", "", regex=True)
+    # Insert spaces between all k-mers
+    inp_dna = inp_dna.apply(lambda x: " ".join([x[i:i+4] for i in range(0, len(x), 4)]))
+
+    # load model to calculate new embeddings
+    tokenizer = PreTrainedTokenizerFast.from_pretrained(model, force_download=True)
+    tokenizer.add_special_tokens({"pad_token": "<UNK>"})
+    bert_model = BertForMaskedLM.from_pretrained(model, force_download=True)
+    embed = bert_model.predic(inp_dna)
+
+    # format lat and lon into coords
+    coords = (inp_lat, inp_lng)
+    # Grab rasters from the tifs
+    ecoLayers = load_dataset("LofiAmazon/Global-Ecolayers")
+    temp = pd.DataFrame([coords, embed], columns = ['coord', 'embeddings'])
+    data = pd.merge(temp, ecoLayers, on='coord', how='left')
+
+    return data
+
 def predict_genus():
-    dna_df = pd.read_csv(dna_file.name)
-    dnaenv_df = pd.read_csv(dnaenv_file.name)
+    data = preprocess()
+    out = xgboost_infer.infer_dna(data)
     
     results = []
-    
-    # envdna_genuses = predict_genus_dna_env(dnaenv_df)
-    # dna_genuses = predict_genus_dna(dna_df)
-    # images = [get_genus_image(genus) for genus in top_5_genuses]
 
     genuses = xgboost_infer.infer()
         
@@ -38,11 +64,11 @@ with gr.Blocks() as demo:
     gr.Markdown("Welcome to Lofi Amazon Beats' DNA Identifier Tool")
 
     with gr.Tab("Genus Prediction"):
-        gr.Markdown("Input a DNA sequence and the coordinates at which its sample was taken to predict the genus of the DNA. Click 'I'm feeling lucky' to see our predictio for a random sequence.")
+        gr.Markdown("Enter a DNA sequence and the coordinates at which its sample was taken to get a genus prediction. Click 'I'm feeling lucky' to see a prediction for a random sequence.")
 
         # Collect inputs for app (DNA and location)
         with gr.Row():
-            inp_dna = gr.Textbox(label="DNA", placeholder="e.g. AACAATGTA... (will be automatically truncated to 660 characters)")
+            inp_dna = gr.Textbox(label="DNA", placeholder="e.g. AACAATGTA... (min 200 and max 660 characters)")
         with gr.Row():
             inp_lat = gr.Textbox(label="Latitude", placeholder="e.g. -3.009083")
             inp_lng = gr.Textbox(label="Longitude", placeholder="e.g. -58.68281")
@@ -57,11 +83,11 @@ with gr.Blocks() as demo:
             gr.Markdown('Make plot or table for Top 5 species')
 
         with gr.Column():
-            genus_out = gr.Dataframe(headers=["DNA", "Coord", "DNA Only Pred Genus", "DNA Only Prob", "DNA & Env Pred Genus", "DNA & Env Prob"])
-            btn_run.click(predict_genus, inputs=[inp_dna, inp_lat, inp_lng], outputs=genus_out)
+            genus_out = gr.Dataframe(headers=["DNA Only Pred Genus", "DNA Only Prob", "DNA & Env Pred Genus", "DNA & Env Prob"])
+            btn_run.click(fn=predict_genus, inputs=[inp_dna, inp_lat, inp_lng], outputs=genus_out)
     
     with gr.Tab('DNA Embedding Space Similarity Visualizer'):
-        gr.Markdown("If the highest genus probability is very low for your DNA sequence, we can still examine the DNA embedding of the sequence in relation to known samples or clues.")
-        
+        gr.Markdown("If the highest genus probability is very low for your DNA sequence, we can still examine the DNA embedding of the sequence in relation to known samples for clues.")
+
 
 demo.launch()

xgboost_infer.py → infer.py

@@ -6,10 +6,16 @@ from sklearn.preprocessing import LabelEncoder
 from datasets import load_dataset
 import pickle
 
+
 def infer_dna(args):
     ecoDf = pd.read_csv(args['input_path'], sep='\t')
     dnaEmbeds = load_dataset("LofiAmazon/BOLD-Embeddings", split='train')
 
+    # load model to calculate new embeddings
+    tokenizer = PreTrainedTokenizerFast.from_pretrained(model, force_download=True)
+    tokenizer.add_special_tokens({"pad_token": "<UNK>"})
+    bert_model = BertForMaskedLM.from_pretrained(model, force_download=True)
+
     modelDNA = load_checkpoint()
     modelDNAEnv = load_checkpoint()
 
@@ -49,6 +55,7 @@ def infer_dna(args):
     y_dna_probs = modelDNAEnv.predict_proba(X_dna)
     DNAEnvGenuses = {}
     for i in range(len()):
+
         topProbs = np.argsort(y_dna_probs[i], axis=1)[:,-3:]
         topClasses = modelDNA.classes_[topProbs]