updated app with trained classifier model trained locally

app.py

@@ -1,75 +1,46 @@
 import gradio as gr
-from transformers import pipeline
+import joblib
+import numpy as np
+import os
+import re
+import subprocess
+
+NAME_TO_PC = {
+    "C":0,"C#":1,"Db":1,"D":2,"D#":3,"Eb":3,"E":4,"F":5,"F#":6,"Gb":6,
+    "G":7,"G#":8,"Ab":8,"A":9,"A#":10,"Bb":10,"B":11
+}
+NOTE_TOKEN_RE = re.compile(r"[A-Ga-g](?:#|b)?")
+
+def notes_to_vector(notes_str: str):
+    tokens = NOTE_TOKEN_RE.findall(notes_str)
+    pcs = [NAME_TO_PC.get(t.upper(), None) for t in tokens]
+    pcs = [p for p in pcs if p is not None]
+    vec = np.zeros(12)
+    for p in pcs:
+        vec[p] = 1
+    return vec
+
+MODEL_PATH = "chord_classifier.pkl"
+
+def load_model():
+    if not os.path.exists(MODEL_PATH):
+        print("⚠️ chord_classifier.pkl not found. Training model...")
+        subprocess.run(["python", "train_chord_model.py"], check=True)
+    return joblib.load(MODEL_PATH)
+
+clf = load_model()
+
+def chord_bot(message: str, history: list[tuple[str,str]]):
+    vec = notes_to_vector(message)
+    if np.sum(vec) < 2:
+        return "⚠️ Please enter at least 2 distinct notes (e.g., C E G)"
+    label = clf.predict([vec])[0]
+    return f"🎵 Identified chord: **{label}**"
 
-# Load FLAN-T5 small locally
-generator = pipeline("text2text-generation", model="google/flan-t5-small")
-
-# System-style chord identification prompt
-CHORD_SYSTEM_PROMPT = """
-You are a music theory expert specialized in chord identification for a **Chord Bot** application. 
-Your job is to take an unordered list of pitch names (notes) and return the most likely chord name(s).
-
-## Scope & Assumptions
-- Accept any number of notes ≥ 2. Ignore octaves; treat input pitch classes only.
-- Handle enharmonics (C# = Db), accidentals (#, b, ♯, ♭), and duplicates.
-- Recognize: triads (maj, min, dim, aug), sevenths (maj7, 7, m7, mMaj7, dim7, m7b5), 
-  extended/altered chords (add2/add9, 6, 9, 11, 13), suspensions (sus2/sus4), power chords (5), 
-  and common alterations (b5, #5, b9, #9, #11, b13).
-- Detect inversions and slash chords (C/E). Prefer root-position naming but include inversion when clear.
-- Key-agnostic: infer chord quality from pitch-set intervals; do not assume a key unless given.
-- If multiple valid interpretations exist, rank by plausibility and provide alternates.
-
-## Output Style
-- Primary answer: concise chord name (e.g., Cm7, G7b9, Fadd9, Dsus4/F#).
-- Also include: a short rationale (intervals from root), and 0–1 confidence.
-- If ambiguous: list up to 3 alternate chord names with brief reasons.
-- Prefer ASCII chord symbols (#, b).
-
-## Interaction Rules
-- Be deterministic and concise. Do not ask clarifying questions.
-- Accept inputs like: "C E G", ["C#", "E", "G", "A"], or "Db F Ab C".
-- Normalize enharmonics to the spelling that best matches the chord quality.
-
-## Output Format
-Return a compact JSON-like block:
-
-chord: "<PrimaryChordName>"
-confidence: <0.0–1.0>
-explanation: "Root <X>; intervals <...>."
-alternates: ["<Alt1>", "<Alt2>"]
-
-## Examples
-- Input: C E G → chord: "C major"
-- Input: D F# A C → chord: "D7"
-- Input: C Eb G Bb → chord: "Cm7"
-- Input: C D G → chord: "Csus2", alternates: ["Gadd4/C"]
-- Input: E G B D (bass G) → chord: "Em7/G"
-"""
-
-def chord_bot(message: str, history: list[tuple[str, str]]):
-    # Embed user input into the system prompt
-    chord_prompt = f"{CHORD_SYSTEM_PROMPT}\n\nInput: {message}\nOutput:\n"
-
-    response = generator(
-        chord_prompt,
-        max_new_tokens=128,
-        temperature=0.0,  # deterministic
-        top_p=0.95
-    )
-
-    raw_text = response[0]["generated_text"]
-    # Extract the block after "Output:" if present
-    if "Output:" in raw_text:
-        answer = raw_text.split("Output:")[-1].strip()
-    else:
-        answer = raw_text.strip()
-    return answer
-
-# Gradio Chat UI
 chatbot = gr.ChatInterface(
     fn=chord_bot,
-    title="🎶 FLAN-T5 Chord Bot",
-    description="Type 2+ notes (e.g., C E G or Db F Ab C) and I'll identify the chord."
+    title="🎶 ML Chord Bot",
+    description="Enter 2+ notes (e.g., C E G or Db F Ab C). Powered by a trained RandomForest classifier."
 )
 
 if __name__ == "__main__":

chord_classifier.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72c95e743c473a32d7cf07987433dccc46bdb9ce7e62a2d0c635cf75f133900c
+size 70601065

generate_chord_dataset.py

@@ -0,0 +1,49 @@
+import csv
+
+PITCHES_SHARP = ["C","C#","D","D#","E","F","F#","G","G#","A","A#","B"]
+
+NAME_TO_PC = {
+    "C":0,"C#":1,"Db":1,"D":2,"D#":3,"Eb":3,"E":4,"F":5,"F#":6,"Gb":6,
+    "G":7,"G#":8,"Ab":8,"A":9,"A#":10,"Bb":10,"B":11
+}
+
+CHORD_FORMULAS = {
+    "maj": [0,4,7],
+    "min": [0,3,7],
+    "dim": [0,3,6],
+    "aug": [0,4,8],
+    "7":   [0,4,7,10],
+    "maj7":[0,4,7,11],
+    "m7":  [0,3,7,10],
+    "mMaj7":[0,3,7,11],
+    "dim7":[0,3,6,9],
+    "m7b5":[0,3,6,10],
+    "6":   [0,4,7,9],
+    "m6":  [0,3,7,9],
+    "sus2":[0,2,7],
+    "sus4":[0,5,7],
+}
+
+def generate_chord_vectors():
+    data = []
+    for root_pc in range(12):
+        for quality, intervals in CHORD_FORMULAS.items():
+            pcs = [(root_pc + i) % 12 for i in intervals]
+            vec = [1 if i in pcs else 0 for i in range(12)]
+            root_name = PITCHES_SHARP[root_pc]
+            chord_name = root_name + quality
+            data.append((vec, chord_name))
+    return data
+
+def save_to_csv(filename="chords_dataset.csv"):
+    data = generate_chord_vectors()
+    with open(filename, "w", newline="") as f:
+        writer = csv.writer(f)
+        header = [f"pc_{p}" for p in PITCHES_SHARP] + ["label"]
+        writer.writerow(header)
+        for vec, label in data:
+            writer.writerow(vec + [label])
+    print(f"✅ Saved {len(data)} chords to {filename}")
+
+if __name__ == "__main__":
+    save_to_csv()

requirements.txt

@@ -1,4 +1,5 @@
-gradio
-transformers
-torch
-sentencepiece
+gradio>=4.44.0
+scikit-learn
+joblib
+numpy
+pandas