Added Ablation Studies

- No Phase Encoding
- No Dendritic Routing
- No Plasticity
- Full [Having Phase Encoding + Dendritic Routing + Plasticity]

-- For Dataset --
1. Using MountainCarv0, Pendulumv1, Cartpolev1, Acrobotv1
2. Using the method of taking the randomness more in compare to the expert and medium one...
3. A1, B1, F1 are the files where we did the: generate -> process -> verify the datasets...

Architecture is still the same, now talking for the scripts and computes:
4 environments x 5 models = 20 scripts will be run in the T4 GPU in Google Colab.

Co-Authored-By: Aditi <76613793+aditi0x@users.noreply.github.com>

Author: Vishal-sys-code

ablation_studies/README.md

@@ -0,0 +1,96 @@
+# Phase 2: Ablation Study & Baseline Comparison
+
+This directory contains the code and resources for the Phase 2 ablation study of the SNN-DT project, which now includes a comprehensive comparison against baseline models.
+
+## 1. Installation
+
+Ensure you have the required dependencies installed from the main project's `requirements.txt`:
+
+```bash
+pip install -r ../requirements.txt
+```
+
+You will also need to install `stable-baselines3`, `pyyaml`, `tqdm`, and `gymnasium`:
+
+```bash
+pip install stable-baselines3 pyyaml tqdm gymnasium
+```
+
+## 2. Dataset Generation
+
+The new dataset generation process is a multi-stage pipeline that produces high-quality, return-stratified datasets.
+
+### Step 1: Generate Raw Trajectories
+
+First, generate the raw trajectories from random, medium, and expert policies.
+
+```bash
+python scripts/A1_generate_trajectories.py
+```
+
+This will save the raw trajectories to `ablation_studies/datasets/raw`.
+
+### Step 2: Process Datasets
+
+Next, process the raw trajectories to create the final, stratified datasets.
+
+```bash
+python scripts/B1_process_datasets.py
+```
+
+This will create `stratified_dataset.npz` and `random_heavy_dataset.npz` in `ablation_studies/datasets/processed` for each environment.
+
+### Step 3: Verify Datasets
+
+Finally, verify the quality of the generated datasets.
+
+```bash
+python scripts/F1_verify_datasets.py
+```
+
+This will generate distribution plots in `ablation_studies/datasets/verification_plots` and print a spike sanity check to the console.
+
+## 3. Run Experiments
+
+To run an experiment, use the `run_experiment.py` script with the desired variant, environment, and seed.
+
+### Ablation Variants
+
+**Example:**
+
+```bash
+python run_experiment.py --variant full --env CartPole-v1 --seed 1001
+```
+
+### Baseline Models
+
+**Example:**
+
+```bash
+python run_experiment.py --variant dt --env CartPole-v1 --seed 1001
+```
+
+### Full Experimental Run
+
+You can run all experiments using a simple shell loop:
+
+```bash
+for variant in full no_phase no_routing no_plasticity dt snn_dt iql cql; do
+  for env in CartPole-v1 Acrobot-v1 Pendulum-v1; do
+    for seed in 1001 1002 1003; do
+      echo "--- Running $variant on $env with seed $seed ---"
+      python run_experiment.py --variant "$variant" --env "$env" --seed "$seed"
+    done
+  done
+done
+```
+
+## 4. Post-process Results
+
+After the experiments are complete, you can generate the plots and summary tables using the `post_process.py` script.
+
+```bash
+python scripts/post_process.py
+```
+
+This will save the figures to the `ablation_studies/figures` directory.

ablation_studies/configs/phase2/cql.yaml

@@ -0,0 +1,2 @@
+model:
+  name: cql

ablation_studies/configs/phase2/dt.yaml

@@ -0,0 +1,2 @@
+model:
+  name: dt

ablation_studies/configs/phase2/full.yaml

@@ -0,0 +1,6 @@
+phase_encoder:
+  enabled: true
+routing:
+  enabled: true
+local_plasticity:
+  enabled: true

ablation_studies/configs/phase2/iql.yaml

@@ -0,0 +1,2 @@
+model:
+  name: iql

ablation_studies/configs/phase2/no_phase.yaml

@@ -0,0 +1,6 @@
+phase_encoder:
+  enabled: false
+routing:
+  enabled: true
+local_plasticity:
+  enabled: true

ablation_studies/configs/phase2/no_plasticity.yaml

@@ -0,0 +1,6 @@
+phase_encoder:
+  enabled: true
+routing:
+  enabled: true
+local_plasticity:
+  enabled: false

ablation_studies/configs/phase2/no_routing.yaml

@@ -0,0 +1,6 @@
+phase_encoder:
+  enabled: true
+routing:
+  enabled: false
+local_plasticity:
+  enabled: true

ablation_studies/configs/phase2/snn_dt.yaml

@@ -0,0 +1,2 @@
+model:
+  name: snn_dt

ablation_studies/experiment_contract.yaml

@@ -0,0 +1,35 @@
+envs: ["CartPole-v1", "Acrobot-v1", "Pendulum-v1", "MountainCar-v0"]
+datasets_per_env:
+  steps: 10000
+  mix: "50% expert / 50% random"
+sequence_length_N: 20
+spiking_window_T: 10
+hidden_dim_d: 128
+num_layers_L: 2
+num_heads_H: 4
+batch_size: 64
+optimizer: AdamW
+lr: 3e-4
+weight_decay: 1e-2
+epochs: 50
+local_lr_eta_local: 0.05
+surrogate_slope_k: 10
+spike_energy_pJ: 5.0
+seeds: [1001, 1002, 1003]
+log_interval_steps: 250
+checkpoint_interval_epochs: 5
+eval_rollouts: 50
+
+# --- Model-specific configs ---
+iql:
+  tau: 0.005
+  temperature: 3.0
+  expectile: 0.7
+  hidden_size: 256
+cql:
+  tau: 0.005
+  temperature: 1.0
+  hidden_size: 256
+  with_lagrange: false
+  cql_weight: 1.0
+  target_action_gap: 10.0