app.py

# -*- coding: utf-8 -*-
"""app.py

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1QIEwA7FDPNIgdUKfLyRF4K3Im9CjkadN

Logistic Map Equation: x
n+1
​
 =r⋅x
n
​
 ⋅(1−x
n
​
 )

- x_n is the current state (a number between 0 and 1).

- x_{n+1} is the next value in the sequence.

- r is the growth rate parameter.

This block:

- Introduces the logistic map function

- Lets us generate sequences with different r values

- Plots them to visually understand convergence, cycles, and chaos
"""

import numpy as np
import matplotlib.pyplot as plt
import random

# Define the logistic map function
def logistic_map(x0: float, r: float, n: int = 100) -> np.ndarray:
    """
    Generates a logistic map sequence.

    Args:
        x0 (float): Initial value (between 0 and 1).
        r (float): Growth rate parameter (between 0 and 4).
        n (int): Number of time steps.

    Returns:
        np.ndarray: Sequence of logistic map values.
    """
    seq = np.zeros(n)
    seq[0] = x0
    for i in range(1, n):
        seq[i] = r * seq[i - 1] * (1 - seq[i - 1])
    return seq

# Plot logistic map sequences for different r values
def plot_logistic_map_examples(x0: float = 0.51, n: int = 100):
    """
    Plots logistic map sequences for several r values to visualize behavior.

    Args:
        x0 (float): Initial value.
        n (int): Number of iterations.
    """
    r_values = [2.5, 3.2, 3.5, 3.9, 4.0]
    plt.figure(figsize=(12, 8))

    for i, r in enumerate(r_values, 1):
        x0_safe = random.uniform(0.11, 0.89)
        seq = logistic_map(x0, r, n)
        plt.subplot(3, 2, i)
        plt.plot(seq, label=f"r = {r}")
        plt.title(f"Logistic Map (r = {r})")
        plt.xlabel("Time Step")
        plt.ylabel("x")
        plt.grid(True)
        plt.legend()

    plt.tight_layout()
    plt.show()

# 🔍 Run the plot function to see different behaviors
plot_logistic_map_examples()

"""- Low r (e.g., 2.5) = stable

- Mid r (e.g., 3.3) = periodic

- High r (e.g., 3.8 – 4.0) = chaotic

Generate synthetic sequences using random r values

Label each sequence as:

- 0 = stable (low r)

- 1 = periodic (mid r)

- 2 = chaotic (high r)

Create a full dataset we can later feed into a classifier
"""

import random
from typing import Tuple, List

# Label assignment based on r value
def label_from_r(r: float) -> int:
    """
    Assigns a regime label based on the value of r.

    Args:
        r (float): Growth rate.

    Returns:
        int: Label (0 = stable, 1 = periodic, 2 = chaotic)
    """
    if r < 3.0:
        return 0  # Stable regime
    elif 3.0 <= r < 3.57:
        return 1  # Periodic regime
    else:
        return 2  # Chaotic regime

# Create one labeled sequence
def generate_labeled_sequence(n: int = 100) -> Tuple[np.ndarray, int]:
    """
    Generates a single logistic map sequence and its regime label.

    Args:
        n (int): Sequence length.

    Returns:
        Tuple: (sequence, label)
    """
    r = round(random.uniform(2.5, 4.0), 4)
    x0 = random.uniform(0.1, 0.9)
    sequence = logistic_map(x0, r, n)
    label = label_from_r(r)
    return sequence, label

# Generate a full dataset
def generate_dataset(num_samples: int = 1000, n: int = 100) -> Tuple[np.ndarray, np.ndarray]:
    """
    Generates a dataset of logistic sequences with regime labels.

    Args:
        num_samples (int): Number of sequences to generate.
        n (int): Length of each sequence.

    Returns:
        Tuple[np.ndarray, np.ndarray]: X (sequences), y (labels)
    """
    X, y = [], []

    for _ in range(num_samples):
        sequence, label = generate_labeled_sequence(n)
        X.append(sequence)
        y.append(label)

    return np.array(X), np.array(y)

# Example: Generate small dataset and view label counts
X, y = generate_dataset(num_samples=500, n=100)

# Check class distribution
import collections
print("Label distribution:", collections.Counter(y))

"""Used controlled r ranges to simulate different market regimes

Created 500 synthetic sequences (X) and regime labels (y)

Now we can visualize, split, and train on this dataset

Visualize:

- Randomly samples from X, y

- Plots sequences grouped by class (0 = stable, 1 = periodic, 2 = chaotic)

Helps us verify that the labels match the visual behavior
"""

import matplotlib.pyplot as plt
import numpy as np

# Helper: Plot N random sequences for a given class
def plot_class_samples(X: np.ndarray, y: np.ndarray, target_label: int, n_samples: int = 5):
    """
    Plots sample sequences from a specified class.

    Args:
        X (np.ndarray): Dataset of sequences.
        y (np.ndarray): Labels (0=stable, 1=periodic, 2=chaotic).
        target_label (int): Class to visualize.
        n_samples (int): Number of sequences to plot.
    """
    indices = np.where(y == target_label)[0]
    chosen = np.random.choice(indices, n_samples, replace=False)

    plt.figure(figsize=(12, 6))
    for i, idx in enumerate(chosen):
        plt.plot(X[idx], label=f"Sample {i+1}")

    regime_name = ["Stable", "Periodic", "Chaotic"][target_label]
    plt.title(f"{regime_name} Regime Samples (Label = {target_label})")
    plt.xlabel("Time Step")
    plt.ylabel("x")
    plt.grid(True)
    plt.legend()
    plt.show()

# View class 0 (stable)
plot_class_samples(X, y, target_label=0)

# View class 1 (periodic)
plot_class_samples(X, y, target_label=1)

# View class 2 (chaotic)
plot_class_samples(X, y, target_label=2)

"""Stable: Sequences that flatten out

Periodic: Repeating waveforms (2, 4, 8 points)

Chaotic: No repeating pattern, jittery

Each of these sequences looks completely different — even though they're all generated by the same equation.

No fixed pattern. No periodic rhythm. Just deterministic unpredictability.

But it's not random — it's chaotic: sensitive to initial conditions, governed by internal structure (nonlinear dynamics).

Split X, y into training and testing sets

Normalize (optional, but improves convergence)

Convert to PyTorch tensors

Create DataLoaders for training
"""

import torch
from torch.utils.data import TensorDataset, DataLoader
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

# Step 1: Split the dataset
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, stratify=y, random_state=42
)

# Step 2: Normalize sequences (standardization: mean=0, std=1)
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)   # Fit only on train
X_test_scaled = scaler.transform(X_test)

# Step 3: Convert to PyTorch tensors
X_train_tensor = torch.tensor(X_train_scaled, dtype=torch.float32)
y_train_tensor = torch.tensor(y_train, dtype=torch.long)

X_test_tensor = torch.tensor(X_test_scaled, dtype=torch.float32)
y_test_tensor = torch.tensor(y_test, dtype=torch.long)

# Step 4: Create TensorDatasets and DataLoaders
batch_size = 64

train_dataset = TensorDataset(X_train_tensor, y_train_tensor)
test_dataset = TensorDataset(X_test_tensor, y_test_tensor)

train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=batch_size)

"""This CNN will:

- Take a 1D time series (length 100)

- Apply temporal convolutions to learn patterns

- Use global pooling to summarize features

- Output one of 3 regime classes
"""

import torch.nn as nn
import torch.nn.functional as F

# 1D CNN model for sequence classification
class ChaosCNN(nn.Module):
    def __init__(self, input_length=100, num_classes=3):
        super(ChaosCNN, self).__init__()

        # Feature extractors
        self.conv1 = nn.Conv1d(in_channels=1, out_channels=32, kernel_size=5, padding=2)
        self.bn1 = nn.BatchNorm1d(32)

        self.conv2 = nn.Conv1d(in_channels=32, out_channels=64, kernel_size=5, padding=2)
        self.bn2 = nn.BatchNorm1d(64)

        # Global average pooling
        self.global_pool = nn.AdaptiveAvgPool1d(1)  # Outputs shape: (batch_size, channels, 1)

        # Final classifier
        self.fc = nn.Linear(64, num_classes)

    def forward(self, x):
        # x shape: (batch_size, sequence_length)
        x = x.unsqueeze(1)  # Add channel dim (batch_size, 1, sequence_length)

        x = F.relu(self.bn1(self.conv1(x)))  # (batch_size, 32, seq_len)
        x = F.relu(self.bn2(self.conv2(x)))  # (batch_size, 64, seq_len)

        x = self.global_pool(x).squeeze(2)   # (batch_size, 64)
        out = self.fc(x)                     # (batch_size, num_classes)
        return out

"""Conv1d:	Extracts local patterns across the time dimension

BatchNorm1d:	Stabilizes training and speeds up convergence

AdaptiveAvgPool1d:	Summarizes the sequence into global stats

Linear:	Final decision layer for 3-class classification
"""

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = ChaosCNN().to(device)

# Define loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)

from sklearn.metrics import accuracy_score, classification_report, confusion_matrix
import seaborn as sns
import matplotlib.pyplot as plt

# Training function
def train_model(model, train_loader, test_loader, criterion, optimizer, device, epochs=15):
    train_losses, test_accuracies = [], []

    for epoch in range(epochs):
        model.train()
        running_loss = 0.0

        for X_batch, y_batch in train_loader:
            X_batch, y_batch = X_batch.to(device), y_batch.to(device)

            optimizer.zero_grad()
            outputs = model(X_batch)
            loss = criterion(outputs, y_batch)
            loss.backward()
            optimizer.step()

            running_loss += loss.item() * X_batch.size(0)

        avg_loss = running_loss / len(train_loader.dataset)
        train_losses.append(avg_loss)

        # Evaluation after each epoch
        model.eval()
        all_preds, all_labels = [], []

        with torch.no_grad():
            for X_batch, y_batch in test_loader:
                X_batch = X_batch.to(device)
                outputs = model(X_batch)
                preds = outputs.argmax(dim=1).cpu().numpy()
                all_preds.extend(preds)
                all_labels.extend(y_batch.numpy())

        acc = accuracy_score(all_labels, all_preds)
        test_accuracies.append(acc)

        print(f"Epoch {epoch+1}/{epochs} - Loss: {avg_loss:.4f} - Test Accuracy: {acc:.4f}")

    return train_losses, test_accuracies

# Train the model
train_losses, test_accuracies = train_model(
    model, train_loader, test_loader, criterion, optimizer, device, epochs=15
)

plt.figure(figsize=(12, 4))

plt.subplot(1, 2, 1)
plt.plot(train_losses, label="Train Loss")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.title("Training Loss Over Time")
plt.grid(True)

plt.subplot(1, 2, 2)
plt.plot(test_accuracies, label="Test Accuracy", color='green')
plt.xlabel("Epoch")
plt.ylabel("Accuracy")
plt.title("Test Accuracy Over Time")
plt.grid(True)

plt.tight_layout()
plt.show()

# Final performance evaluation
model.eval()
y_true, y_pred = [], []

with torch.no_grad():
    for X_batch, y_batch in test_loader:
        X_batch = X_batch.to(device)
        outputs = model(X_batch)
        preds = outputs.argmax(dim=1).cpu().numpy()
        y_pred.extend(preds)
        y_true.extend(y_batch.numpy())

# Confusion matrix
cm = confusion_matrix(y_true, y_pred)
labels = ["Stable", "Periodic", "Chaotic"]

plt.figure(figsize=(6, 5))
sns.heatmap(cm, annot=True, fmt="d", cmap="Blues", xticklabels=labels, yticklabels=labels)
plt.title("Confusion Matrix")
plt.xlabel("Predicted")
plt.ylabel("Actual")
plt.show()

# Classification report
print(classification_report(y_true, y_pred, target_names=labels))

"""Input an r value (between 2.5 and 4.0)

Generate a logistic map sequence

Feed it to your trained model

Predict the regime

Plot the sequence and overlay the prediction
"""

# Label map for decoding
label_map = {0: "Stable", 1: "Periodic", 2: "Chaotic"}

def predict_regime(r_value: float, model, scaler, device, sequence_length=100):
    """
    Generates a logistic sequence for a given r, feeds to model, and predicts regime.
    """
    assert 2.5 <= r_value <= 4.0, "r must be between 2.5 and 4.0"

    # Generate sequence
    x0 = np.random.uniform(0.1, 0.9)
    sequence = logistic_map(x0, r_value, sequence_length).reshape(1, -1)

    # Standardize using training scaler
    sequence_scaled = scaler.transform(sequence)

    # Convert to tensor
    sequence_tensor = torch.tensor(sequence_scaled, dtype=torch.float32).to(device)

    # Model inference
    model.eval()
    with torch.no_grad():
        output = model(sequence_tensor)
        pred_class = torch.argmax(output, dim=1).item()

    # Plot
    plt.figure(figsize=(10, 4))
    plt.plot(sequence.flatten(), label=f"r = {r_value}")
    plt.title(f"Predicted Regime: {label_map[pred_class]} (Class {pred_class})")
    plt.xlabel("Time Step")
    plt.ylabel("x")
    plt.grid(True)
    plt.legend()
    plt.show()

    return label_map[pred_class]

predict_regime(2.6, model, scaler, device)
predict_regime(3.3, model, scaler, device)
predict_regime(3.95, model, scaler, device)

import yfinance as yf
import numpy as np
import torch
import matplotlib.pyplot as plt
import gradio as gr

# --- Label map used by your model ---
label_map = {
    0: "Stable",
    1: "Periodic",
    2: "Chaotic"
}

# --- Fetch + Predict Function ---
def fetch_and_predict(ticker: str, interval: str, model, scaler, device):
    try:
        ticker = ticker.strip().upper()

        # Ensure enough data is fetched based on interval
        def get_valid_period(interval):
            if interval in ["1m", "2m", "5m", "15m", "30m", "60m", "90m"]:
                return "60d"   # yfinance limit for intraday
            elif interval == "1d":
                return "1y"
            elif interval == "1wk":
                return "2y"
            elif interval == "1mo":
                return "5y"
            else:
                return "1y"

        period = get_valid_period(interval)

        # Pull price data
        df = yf.download(ticker, period=period, interval=interval, progress=False)

        if df.empty or 'Close' not in df:
            return None, f"❌ Failed to load data for: {ticker} with interval {interval}"

        prices = df['Close'].dropna().values
        if len(prices) < 101:
            return None, f"⚠️ Not enough price points (only {len(prices)}) for interval: {interval}"

        # Compute log returns
        log_returns = np.diff(np.log(prices))
        log_returns = log_returns[~np.isnan(log_returns)]
        log_returns = log_returns[~np.isinf(log_returns)]

        if len(log_returns) < 100:
          pad_size = 100 - len(log_returns)
          padded_seq = np.pad(log_returns, (pad_size, 0), mode='constant', constant_values=np.mean(log_returns))
        else:
          padded_seq = log_returns[-100:]

        # Prepare input
        seq = log_returns[-100:].reshape(1, -1)
        seq_scaled = scaler.transform(padded_seq.reshape(1, -1))
        seq_tensor = torch.tensor(seq_scaled, dtype=torch.float32).to(device)

        # Model prediction
        model.eval()
        with torch.no_grad():
            output = model(seq_tensor)
            pred_class = torch.argmax(output, dim=1).item()
            regime = label_map[pred_class]

        # Plotting
        fig, axs = plt.subplots(2, 1, figsize=(8, 5))
        axs[0].plot(prices, label="Price")
        axs[0].set_title(f"{ticker.upper()} Price")
        axs[0].grid(True)

        axs[1].plot(log_returns, color='orange', label="Log Returns")
        axs[1].axhline(0, linestyle='--', color='grey', alpha=0.6)
        axs[1].set_title("Log Returns")
        axs[1].grid(True)

        plt.tight_layout()

        if pad_size > 0:
          return fig, f"Predicted Regime: {regime} (⚠️ Sequence padded with mean returns: {pad_size} points)"

    except Exception as e:
        return None, f"❌ Error: {str(e)}"

# --- Wrapper for Gradio Interface ---
def real_data_interface(ticker, interval):
    fig, result = fetch_and_predict(ticker, interval, model, scaler, device)
    if fig is None:
        return None, result
    return fig, result

# --- Real Data Interface Block ---
real_interface = gr.Interface(
    fn=real_data_interface,
    inputs=[
        gr.Textbox(label="Enter Stock Ticker (e.g., AAPL, TSLA)"),
        gr.Dropdown(
            choices=["1d", "1wk", "1mo", "30m", "60m"],
            value="1d",
            label="Select Interval"
        )
    ],
    outputs=[
        gr.Plot(label="Price + Return Sequence"),
        gr.Label(label="Predicted Regime")
    ],
    title="📈 Real-World Market Regime Classifier",
    description="Fetches stock data using yfinance and classifies the regime as Stable, Periodic, or Chaotic using your CNN model."
)

# --- Simulated Chaos Interface (assuming already defined elsewhere) ---
# interface = ...

# --- Final Gradio App with Tabs ---
gr.TabbedInterface(
    interface_list=[real_interface],
    tab_names=["🔬 Simulated Chaos (r)", "📈 Real Market Data"]
).launch()